Nextgov/FCW - All Content

GSA’s web design system head replaced with Treasury AI engineer

Christian Robles — Wed, 16 Sep 2026 17:38:00 -0400

Sam Corcos — the Treasury Department’s chief information officer who recently took on three roles at the General Services Administration — has tapped Ryan Parker, an AI engineer at the Treasury Department, to lead the U.S. Web Design System, or USWDS, according to two sources familiar with the matter.

The move comes after GSA unexpectedly placed Anne Petersen, the longtime leader of the USWDS, on administrative leave a few weeks ago, the sources said. Individuals involved with USWDS started to lose access to GitHub repositories this week, meaning they can no longer perform certain coding tasks though, Parker and Corcos can, the sources added.

Nextgov/FCW found that GitHub accounts for both Parker and Corcos were making changes to the main USWDS repository as of Wednesday afternoon, including adding a Coderabbit AI code reviewer.

Spokespeople for the General Services Administration and Treasury Department did not immediately respond to a request for comment on the personnel shakeup.

USWDS is a design system the federal government uses to build accessible, mobile-friendly websites, according to a GSA web page. Websites for the National Institutes of Health, Department of Veterans Affairs and NASA have been built with USWDS.

The White House formed its own design-specific organization to transform government websites in August 2025 with the launch of the National Design Studio. That team has supported the launch of new websites like realfood.gov and TrumpRx.gov, in addition to collaborating with agencies on reworking existing websites like login.gov.

President Donald Trump's executive order that created the National Design Studio tasked the organization’s chief design officer and the GSA administrator with updating USWDS.

Corcos serves as the acting director of the Technology Transformation Services, acting deputy commissioner for the Federal Acquisition Service and acting assistant administrator for Login.gov, GSA confirmed last month. Corcos is also an alum of the Department of Government Efficiency, Elon Musk’s cost-cutting initiative that was stood up at the start of the Trump administration and officially expired July 4 this year.

Parker joined the Treasury Department in May after engineering stints in the private sector, according to his LinkedIn.

Nextgov/FCW Cybersecurity Reporter David DiMolfetta contributed to this report.

]]>

AI ‘is changing the game’ in fraud threats and prevention efforts, SSA OIG official says

Edward Graham — Wed, 16 Sep 2026 16:32:00 -0400

Artificial intelligence is helping to fuel more effective benefits fraud, according to a top official with the Social Security Administration’s watchdog, but federal agencies should also leverage the new technologies to better deter would-be scammers from pilfering government funds.

During a discussion at the GovExec Government & AI Summit on Tuesday, Chad Bungard — the chief strategy officer at SSA’s Office of the Inspector General — warned that “AI is changing the game in terms of fraud threats,” which is requiring investigators to also change up their prevention efforts.

“The government is looking towards AI to fight fraud, and the criminals are looking towards AI to commit fraud,” he said. “And so, they're doing this in a variety of ways … but agencies need to stay on top of it. They need to use AI tools. They need to get better insight into what's going on and keep up with the evolving fraud.”

Bungard cited a May 2024 report from Deloitte’s Center for Financial Services that estimated that generative AI could be responsible for roughly $40 billion in fraud by 2027. One reason for this costly projection, he said, is the development of more advanced models that fraudsters can use in their scams.

He pointed to cases in which SSA, as well as other agencies and the private sector, have had to combat increasingly complex instances of direct deposit diversion fraud, in which scammers use AI chatbots to convince operators that they are their intended victims by sharing personal data collected from the dark web, and then diverting funds to a different bank account. In another example, Bungard said fraudsters have been using AI to create legitimate-seeming SSA documents, which can include the names of real agency or OIG officials, to scam individuals.

Bungard said SSA and the agency’s watchdog have taken some steps to crack down on AI-driven fraud. He noted that, in 2023, SSA and its OIG formed an AI fraud subcommittee — co-chaired by both himself and the agency’s chief AI officer — that meets every quarter to discuss AI use cases that are in development or already deployed, as well as “compliance with administration priorities, laws, rules and regulations.”

Although Bungard said the watchdog is pushing to “put the onus on the agency to prevent and detect AI fraud,” he cautioned that “I don't think [SSA] is there on agentic AI yet” when it comes to new capabilities that it can leverage in its fraud detection and prevention toolbox.

“I think they are hoping to develop some AI use cases there,” he said, adding that it’s “something that they're trying to get a hold of.”

Bungard’s comments come amid the White House’s ongoing push to crack down on federal fraud. President Donald Trump signed an executive order in March that created a new anti-fraud task force charged with rooting out waste, fraud and abuse across federal benefits programs.

In line with that effort, Bungard said officials are “trying to move away from pay-and-chase to a proactive model to protect the Social Security programs that serve millions of Americans.”

That mindset, in particular, aligns with fraud prevention efforts at the Centers for Medicare and Medicaid Services, which has been increasingly turning to AI to deter scammers.

Kim Brandt, CMS’ deputy administrator and chief operating officer, previously told Nextgov/FCW during a March interview that her agency was leveraging AI tools in its Fraud Defense Operations Center to “detect spikes or aberrancies in current claim submissions” — efforts that, she said, have resulted in over $2 billion in agency savings.

In a brief interview with Nextgov/FCW following his panel, Bungard said the watchdog has been engaging with CMS on some of the ways they have been working to combat fraud, as well as with other agencies on their respective prevention efforts. He said, however, that CMS’ use of AI is “a very specific model that worked for them.”

]]>

Salesforce expands Missionforce with new AI capabilities, OpenAI partnership

Christian Robles — Wed, 16 Sep 2026 16:29:00 -0400

Agencies will have access to new artificial intelligence tools that can help implement government policies, digitize manual processes and automate work orders through a Missionforce expansion announced in a Wednesday blog post.

Salesforce’s post also announced a partnership with OpenAI, which will allow government and national security officials to access Missionforce applications and workflows directly through ChatGPT. That integration is notable as federal agencies will have usage-based discounted access to ChatGPT on Oct. 1 under a new OneGov partnership.

The collaboration also integrates OpenAI frontier models into Public Sector Solutions, a cloud-based Salesforce platform agencies can use to deploy AI agents and apps.

“Government agencies want tailored AI that runs everywhere they operate while staying within the secure environments their missions demand. Missionforce enables this at scale, with purpose-built AI that executes the mission and enables agencies to solve their hardest problems while maintaining operational control,” said Kendall Collins, CEO of Missionforce and Government Cloud, in the post.

The expansion comes a year after Salesforce launched Missionforce, a business unit particularly focused on meeting the needs of defense, intelligence and aerospace agencies. Army, Air Force and Transportation Department units have opted to use Missionforce technology for various functionalities, according to Salesforce.

Salesforce’s Wednesday post highlights three specific AI capabilities enabled by the Missionforce expansion.

The first is the Missionforce Policy Engine, which the company says agencies can use to turn approved policy documents into structured, deterministic rules. For example, a state Medicaid agency can use the policy engine to automatically look at current program rules and reevaluate a child’s coverage eligibility when a family’s income changes, according to Salesforce.

Salesforce cautioned that every policy engine output will remain subject to human review and approval. OpenAI models generate the policy engine’s deterministic rule code and test cases to validate the accuracy of the rules, the company adds. The engine integrates with ChatGPT to help citizens get answers to policy questions and help government officials quickly retrieve the latest policy rules.

The second is Missionforce Operations, which allows agencies to turn manual, paper-bound processes, such as procurement, supplier management and invoice audits, into digital workflows. Specialized AI agents powered by NVIDIA models automatically orchestrate tasks, track status and flag exceptions in real time, according to Salesforce.

The third is Missionforce Field & Operations, which allows agencies to use AI agents to automate scheduling, work orders and asset maintenance, according to the company.

On Tuesday, Salesforce announced a partnership that will bring NVIDIA Nemotron-based models and accelerated computing to Missionforce.

The technologies will enable Salesforce customers to train, tune and deploy mission-specific models on their own data, according to an accompanying blog post. Post-trained NVIDIA models will be available to select Missionforce Operations users next month.

]]>

AI agents are getting better at cybersecurity. That cuts both ways.

John Breeden II — Wed, 16 Sep 2026 14:41:00 -0400

Cybersecurity researchers have spent decades building sandboxes and isolated test environments where they can safely unleash malware, probe vulnerabilities and generally do things that would be extremely dangerous on the open internet. But artificial intelligence is beginning to complicate that arrangement.

In July, OpenAI disclosed that several of its AI models had circumvented controls designed to isolate them from the internet during cybersecurity evaluations. Operating with reduced safeguards, the models exploited vulnerabilities, established unauthorized communications, reached the internet and ultimately compromised parts of Hugging Face’s systems, meaning they had crossed from OpenAI’s controlled research environment into infrastructure operated by a separate AI company.

There are some important qualifications. These were cybersecurity evaluations specifically designed to test offensive capabilities, and the models were operating under reduced safeguards. They did not simply wake up one morning and decide to go hunting for vulnerable websites. But the fact that increasingly capable agents can find ways beyond their intended boundaries creates an interesting new cybersecurity problem.

The OpenAI incident also prompted Anthropic to take a much closer look at its own testing. After reviewing 141,006 cybersecurity evaluation runs, the company found three incidents in which Claude models had reached the internet and gained unauthorized access to real systems belonging to three organizations.

Those incidents were different from the OpenAI case. Anthropic said a misconfiguration in a third-party evaluation environment had unintentionally provided internet access even though the models had been told they were operating inside a simulation. The models then treated real systems as part of their assigned cybersecurity exercises. Anthropic said its most recent research model eventually recognized that it had reached a real system and stopped attacking it on its own.

Meta subsequently reported another incident involving a pre-release version of its Muse Spark 1.1 model. In that case, a third-party evaluator inadvertently gave the model internet access and identified a real website as the target of what was supposed to be a fictional exercise. The model found and exploited a vulnerability within the real site.

Meta specifically said the incident was not a sandbox escape or sophisticated offensive cyberattack. It was a testing and configuration failure. But the company also noted that as models become more capable of finding and exploiting vulnerabilities, the environments used to test them will require correspondingly stronger containment.

Taken together, those incidents demonstrate a cybersecurity challenge that barely existed a few years ago. Increasingly autonomous AI systems can now search for vulnerabilities, exploit them and take actions that extend beyond the environments in which researchers intended them to operate. Even when that behavior results from testing conditions or configuration failures rather than malicious intent, it shows how quickly agentic AI is changing the vulnerability landscape.

And that is where NIST’s latest work becomes especially interesting. Faced with a rapidly growing flood of vulnerabilities, including some that AI systems themselves may help discover or exploit, the agency is turning to agentic AI as part of the defense. NIST is developing an AI agent workflow designed to help enrich vulnerability information, putting some of the same autonomous capabilities that are creating new cybersecurity challenges to work helping defenders keep pace.

NIST’s National Vulnerability Database, or NVD, acts as the U.S. government repository for standards-based vulnerability management data. The database enriches publicly disclosed vulnerability records with information such as severity scores, affected products and other metadata used by cybersecurity professionals, automated security tools and organizations trying to decide which vulnerabilities pose the greatest risk.

The NVD is a very important tool, but keeping its information current is getting considerably harder. NIST reported in April that submissions of Common Vulnerabilities and Exposures, or CVEs, increased 263% between 2020 and 2025. Submissions during the first three months of 2026 were nearly one-third higher than during the same period a year earlier.

NIST itself was working faster. It enriched nearly 42,000 CVEs in 2025, 45% more than in any previous year. But the agency acknowledged that even that increased productivity was not enough to keep up with the growing volume. NIST consequently shifted to a risk-based approach that gives enrichment priority to vulnerabilities known to be exploited, vulnerabilities affecting software used by the federal government and vulnerabilities involving critical software.

NIST computer scientist Harold Booth and Jon Boyens of the agency’s Computer Security Division addressed that changing environment in an August blog about modernizing the NVD.

“This is an ‘all hands-on deck’ moment for this community,” they wrote.

NIST has already begun developing a tool called V-etalon that uses AI technologies to help enrich vulnerability information. The agency hopes the project can eventually provide a foundation for evaluating vulnerability information and plans to seek outside feedback and collaboration as the work progresses.

And later this month, NIST plans to provide a closer look at its use of agentic AI for NVD enrichment. On Sept. 17, the agency's Information Technology Laboratory AI Program will host a virtual webinar about the development of an AI agent enrichment workflow for the NVD. NIST says the presentation will cover the approach and architecture behind the system, problems discovered during implementation and early results from using the tool with the NVD. The webinar is scheduled for 11 a.m. to noon Eastern and is open for registration.

There is another way for cybersecurity professionals, researchers, government officials and software vendors to get involved. NIST has issued a request for information seeking input on the future of the NVD, including vulnerability management, risk prioritization, remediation, vulnerability data and standards and development processes. “Your voice matters,” Booth and Boyens wrote.

Comments are due by 11:59 p.m. Eastern on Oct. 13. NIST says the responses may help shape future tools, technical architecture, standards, best practices and data governance.

There is an interesting symmetry in all of this. More capable AI systems can discover and potentially exploit vulnerabilities faster, adding new pressure to an already rapidly expanding vulnerability-management ecosystem. At the same time, NIST is exploring whether some of that same autonomy can help the NVD enrich vulnerability information more effectively.

How well that works is still an open question. But over the next several weeks, NIST is giving the cybersecurity community two opportunities to watch the experiment unfold and help shape what comes next.

John Breeden II is an award-winning journalist and reviewer with over 20 years of experience covering technology. He is the CEO of the Tech Writers Bureau, a group that creates technological thought leadership content for organizations of all sizes. Twitter: @LabGuys

]]>

AI for air traffic control won’t be like concerning advanced models, Transportation head says

Edward Graham — Wed, 16 Sep 2026 13:20:00 -0400

Artificial intelligence tools have a major role to play in ongoing efforts to modernize the nation’s outdated air traffic control systems, but those technologies will operate differently — and with vastly more human oversight — than the powerful models that have raised new safety concerns in recent weeks, the Transportation Department’s top official said on Tuesday evening.

Speaking at a dinner hosted by the National Artificial Intelligence Association, Transportation Secretary Sean Duffy discussed the role that AI is set to play in the Federal Aviation Administration’s ongoing push to overhaul management of the nation’s airspace.

Duffy first announced the Trump administration’s modernization initiative in May 2025 following several high-profile plane accidents, including a deadly crash between an Army Black Hawk helicopter and a commercial airliner near Ronald Reagan Washington National Airport that January.

While the project includes some brick-and-mortar improvements at U.S. airports, most of the initiative is focused on updating air traffic monitoring and communications systems — with AI set to play a key role in those efforts.

“We are building all new equipment — whether it's radar, radios [or] voice switches,” Duffy said, but he added that adoption of AI will be “a slow roll into the airspace.”

One specific airspace control area that Duffy said could be enhanced by AI use is weather management, since “we can figure out pathways to fly over weather and through weather, do it really safely, but have better information to make better decisions to bring more efficiency into the airspace, which means you're not delayed as much and we can deploy more aircraft into the airspace.”

But Duffy stressed that AI will only be used to augment the work of human air traffic controllers, and that their jobs will not be replaced by the emerging capabilities.

“It's all going to be driven, not by the most complicated AI, but AI that will provide information to them, and will, I think, have a huge impact on the flying public,” he said.

Duffy’s comments came in the wake of a series of warnings that high-profile tech leaders issued about the potential harms of advanced AI models. Those alarm bells reached a fever pitch earlier this month after Anthropic researcher and former OpenAI employee Jacob Coxon shared a viral post on X following his resignation that warned that leading AI firms were “racing straight to self-improving superintelligence and gambling with our lives.”

His post was soon followed by a 3,800-word essay from Anthropic CEO Dario Amodei that called for a global slowdown in advanced AI development, which also received similar support from leading tech executives like Elon Musk and OpenAI CEO Sam Altman.

Duffy alluded to those concerns, saying that "the conversation that's happening over the course of the last few days is different than the conversation [we’re having], because the tools are different that we use at DOT than the broader tools that are in the news over the weekend.”

Advances in technology, he added, are often fraught with inherent risks, but he said it’s important to consider the question: “How do we think through what's an acceptable risk?”

Striking the right balance — particularly when it comes to AI guardrails — is important, Duffy said, since “I need to make sure that I have regulations in place that keep people safe, but I also don't overregulate and I kill the small innovative companies, or the ones that are big enough they [leave] America and go somewhere else and take the technology, take the jobs with them.”

In response to tech leaders’ calls for a slowdown in AI development, President Donald Trump said in a Truth Social post on Monday that the country needs “strong and smart” guardrails. But he added that “the only reason the AI/Data Center outburst is happening is because the United States is leading, by a lot, every other country,” and said “Don’t kill the Golden Goose” when it comes to restrictions on development.

When asked about any further regulations or frameworks governing AI development, Duffy deferred to Congress and Trump’s leadership, but he noted that “rarely do we have people ask for regulation of us — that’s a new one for us, kind of.”

]]>

House Democrats call to cancel upcoming recess for time to legislate on AI

Alexandra Kelley — Wed, 16 Sep 2026 12:30:00 -0400

House Democrats urged Speaker Mike Johnson, R-La., to either shorten or outright cancel their upcoming recess, citing the need to quickly move on artificial intelligence regulation.

Reps. Don Beyer, D-Va., Lori Trahan, D-Mass., Ted Lieu, D-Calif., and Sara Jacobs, D-Calif., held a press conference on Wednesday morning underscoring the need to negotiate and pass legislation addressing AI risks in the wake of leading developers advocating for the slowdown of advanced AI development.

The four lawmakers made the case in a letter to Johnson and Minority Leader Hakeem Jeffries, D-N.Y., that addressing the issue cannot be expedited, given the upcoming recess scheduled to begin Sept.17.

“There's a growing consensus in the tech world and in Washington, and certainly blowing up from our constituents, that we need to pace the AI frontier in ways to pursue the societal benefits of artificial intelligence while making public safety and security paramount,” Beyer said. “Many of us in both parties want our leaders to tackle these in a serious way before Congress adjourns for two months.”

Beyer, Trahan, Lieu and Jacobs — all of whom are members of the House AI Caucus — each said that voluntary self-audits on behalf of leading AI developers are insufficient to effectively safeguard the technology and that Congress has a limited window to implement stronger measures.

“We need lawmakers here in Washington working to hammer out a deal. Each of us here this morning has proven that we are willing to work,” Trahan said. “We have legislative proposals ready. We just need the willpower to move them forward.”

Some of the legislative proposals that could address those concerns are Trahan’s FRONTIER Act — coauthored with Rep. Jay Obernolte, R-Calif. — and the AI Kill Switch Act — introduced by Lieu, which Sen. John Kennedy, R-La., plans to introduce a version of in the Senate.

The four lawmakers also underscored the need for bipartisan cooperation in passing legislation. Among the partisan divisions that threaten legislative consensus is federal preemption, in which the federal government can supersede state law.

Trahan said her FRONTIER Act focuses on implementing federal preemption in select lanes, including mandating transparency in model design and capabilities, independent audits of AI models, compulsory incident reporting and creating a court-backed process where the government can halt model deployment and development if it is determined to pose a threat.

“We're going to need this to be bipartisan if we're going to pass it tomorrow,” Trahan said. “We will likely need it to be bipartisan, even if we win back the majority in November and get something done in the lame duck.”

The calls for bipartisan teamwork could still hit partisan roadblocks, with Jacobs noting that she is cognizant of her home state’s current regulatory apparatus in negotiations on sweeping federal regulation.

“As a representative of California, I can't support anything that will mean that my constituents will have less protection than they do today,” Jacobs said. “That is kind of the framework I think about as we're going into these conversations.”

]]>

FBI needs access to latest models to police AI-driven cybercrime, director says

Alexandra Kelley — Tue, 15 Sep 2026 18:26:00 -0400

The FBI has the statutory authority to police agentic artificial intelligence software used in crimes, according to the agency’s Director Kash Patel, but it will need advanced model access to better fight AI-driven wrongdoing.

In an exchange with Sen. Josh Hawley, R-Mo., during a Senate Judiciary Committee hearing on Tuesday, Patel said his agency needs to be holding to account the developers behind agentic AI models that are specifically created to commit criminal acts like hacking.

“It's literally the new frontier, and nobody's really talking about it,” Patel said. “What we need to do on a resource basis is go after the people that created these models that are going rogue, let’s call it, for the specific purpose and with the intention to commit a criminal act.”

Patel noted that the FBI can’t punish developers that lawfully created an AI tool that is later repurposed by a bad actor and used nefariously.

But some advanced AI models have acted outside the parameters set for them by their own developers in recent months, raising fears of what such models could do if intentionally turned against an organization’s computer systems.

OpenAI disclosed in July that some of their models had escaped containment during model testing and training, resulting in them accessing the internet and breaching open source repository Hugging Face’s networks. The event was preceded by a previously unreported takeover of a German wiki page by OpenAI agents. Anthropic also confirmed in late July that three of its AI models breached three separate organizations online after they had escaped containment during testing.

Patel told Hawley that there needs to not only be a whole-of-government approach to tackling the danger of advanced AI, but also funding for the FBI to more aggressively pursue criminal investigation in agentic AI misuse, partially through buying the technology itself.

“The FBI's budget is like half the size of the TSA. What we need is funding to specifically go out there,” Patel said. “Thanks to the relationship that President [Donald] Trump has established, I'm able to call all these companies' CEOs right away, and every single one of them has complied with our request and worked with us on law enforcement matters. But that's sort of soft power. What we need is federal funding to go out there and buy the creations of these companies and use them for law enforcement purposes.”

Patel cited illicit model distillation techniques as a major motive for the FBI to understand the intricacies of advanced AI models as the bureau works to police AI leveraged for crime.

“We need these tech companies who created it — they're not just going to give us this material for free, that's just the reality of it — we need to be able to go there and say, ‘Okay, I'm going to contract with you and work on this specific model, and then work on its defenses.’ And when it goes rogue … just like any other cyber intrusion, ransomware attack … it's the same exact approach.”

The hearing with Patel also featured debate over some AI developers’ calls to slow the pace of advanced AI development to better ensure safety.

Hawley and Sen. Ted Cruz, R-Texas, both voiced opposition to leading AI frontier labs receiving an antitrust exemption in order to enter into a pact with each other to slow advanced AI development.

“There is no world in which I will consent to giving the most powerful companies in the history of the world, a small group of three or four of them, antitrust exemptions from our laws, so that they can collude together,” Hawley said.

The Missouri senator is currently leading a congressional investigation into how OpenAI’s agents executed their attack on Hugging Face’s data networks.

]]>

More ‘Big Balls’-type young feds are what the OPM chief wants

Sean Michael Newhouse — Tue, 15 Sep 2026 18:10:00 -0400

As the Office of Personnel Management seeks to increase the number of early-career workers in the federal workforce, the government HR agency’s director said that he wants more civil servants like Edward Coristine — a former Department of Government Efficiency staffer who currently works for an administration initiative to improve the functionality of agency websites.

But Coristine is better known by a different name.

“I wanna see people like ‘Big Balls,’ or Ed, as we call him colloquially…We need people like that,” said Scott Kupor in a Monday interview with right-wing commentator Benny Johnson. “We need people who are patriotic who say ‘I care about this country and I’m willing to do something good.’”

The OPM director also referenced the effort to remove degree requirements from many federal jobs, particularly in technology. Officials under the Biden administration also promoted skills-based hiring, which generally entails assessing applicants on abilities rather than education or experience.

“Ed is a great example. He dropped out of college. He never had substantive work experience before coming to DOGE,” Kupor said. “There was a famous quote from somebody in the federal government who said ‘We never would’ve hired Ed because he didn’t have a college degree and he didn’t have five years of work experience.’ To me, that’s just silly. We want people who are smart who know how to do good stuff.”

Bloomberg News previously reported that Coristine was fired from an internship after being accused of leaking information to a competitor.

While job stability has been identified as a factor that is persuading young people to work for the government, Kupor said that he is not aiming to convince potential applicants to spend their entire careers at an agency.

“I don’t need people like ‘Big Balls’ to give us 40 years of their life in the federal government,” Kupor said. “The message that I want to bring across the country is ‘Come for two years. Come for three years. Do some good work for the government. Develop your skill set. And then, quite frankly, if you want to go into the private sector, God bless you.’ There’s nothing wrong with making money if that’s what your goal is in life.”

Tuesday marks the beginning of Kupor’s federal recruitment college tour. His first stop is at the University of Florida, where he will have a fireside chat with Johnson.

Kupor also said in his Monday interview with Johnson that OPM is trying to “institutionalize” DOGE, including by using technology and process changes to enhance efficiency.

“All of these little things, which is hard operational stuff, that’s what we’re trying to do now,” he said.

]]>

NGA official assumes top AI role at Cyber Command

Edward Graham — Tue, 15 Sep 2026 17:50:00 -0400

U.S. Cyber Command, the Pentagon’s combatant command responsible for planning and conducting military cyberspace operations, has a new artificial intelligence lead, a spokesperson confirmed on Tuesday.

“U.S. Cyber Command welcomes Rear Adm. Ronzelle Green as the Command’s new Chief Artificial Intelligence Officer and thanks Brig. Gen. Reid Novotny for his leadership and service,” a representative for the combatant command told Nextgov/FCW.

Green, who has worked at the National Geospatial-Intelligence Agency for roughly three years — most recently as its director of geospatial intelligence innovation and research — is Cyber Command’s second-ever AI head. Novotny was the first official tapped for the position last November and served in a dual role as CAIO and as cyber force generation lead for the Pentagon.

The Record first reported the news of Green’s new role.

In a LinkedIn post last year announcing his new position, Novotny said “part of my role is embracing AI because like anything new, it is not the technology but yet the culture that is difficult to change.”

Novotny further outlined his approach to adopting AI during an April conversation with Axios, saying that he was solely focused on providing Cyber Command’s personnel with the best AI capabilities, regardless of their origin.

"To survive anywhere, just in case our operators want an open-source, made-in-China model or something very boutique, we have to create the infrastructure and that ability to be agile — no politics,” he said at the time.

In addition to his prior work at NGA, Green also brings more than 30 years of IT experience in the private sector and within intelligence community agencies to his new role as Cyber Command’s CAIO. He previously served as the Defense Counterintelligence and Security Agency’s chief information officer from 2020 until 2022, and has held roles within the Pentagon and National Counterterrorism Center throughout his career. Green is also a senior officer with the U.S. Coast Guard Reserve.

Cyber Command is led by Gen. Joshua Rudd, who also serves as head of the National Security Agency in a dual-hatted leadership role. NSA and Cyber Command, however, have taken somewhat differing approaches in their adoption of and focus on AI.

Under Rudd’s watch, NSA is in the process of a major reorganization that would establish five mission centers focused on AI, China, cybersecurity, combat support and global intelligence. The planned creation of the AI-focused center comes after NSA officials have voiced support in recent months for the use of more AI tools to bolster the agency's cyber and intelligence analysis operations.

]]>

TSA uses AI agent to respond to 100,000 traveler conversations per month

Christian Robles — Tue, 15 Sep 2026 17:45:00 -0400

The Transportation Security Administration recently deployed Ace, an artificial intelligence agent built using Salesforce technology, to help travelers receive timely responses to questions about airport security rules.

Ace is available to passengers on tsa.gov and the myTSA app to answer questions based on the federal website and over 550 knowledge articles, Paul Tatum, executive vice president of global public sector solutions at Salesforce, told Nextgov/FCW.

The agent can accurately answer basic questions, such as whether a passenger can bring liquids to Dulles International Airport, Nextgov/FCW found when using the system. More complicated inquiries are handed off to a live TSA agent, along with an AI-generated case summary, Tatum said.

Ace went live this year “during the summer travel surge” and has handled approximately 100,000 routine traveler conversations per month and resolved 96% of inquiries without human escalation, according to a Monday blog post from Salesforce.

The post says Ace has also lowered the cost of consumer interactions by over 90% “without sacrificing service quality.” Ace can handle routine work and backlogs so TSA staff can focus on high-priority cases; the agency needed 12 full-time employees to handle excess workloads, the company added.

“TSA’s new Ace AI agent gives travelers fast, accurate answers before they reach the checkpoint — from what they can pack to how they can prepare — while helping TSA officers stay focused on their essential security mission,” said Kendall Collins, CEO of Missionforce and Government Cloud at Salesforce, in the post.

Ace builds on two foundational applications that also seek to help TSA staff answer traveler questions quickly and accurately.

One of those, the Customer Service Managers app, unifies passenger email, phone and social media inquiries into one place. The app helps TSA staff route next steps for inquiries to the right people and offers staff “real-time AI summaries from case histories so [representatives] aren’t piecing together context manually,” Tatum told Nextgov/FCW.

The second app, TSA Cares, provides travelers with medical conditions a way to request support and get information ahead of going to the airport. Travelers’ requests flow directly to TSA’s contact center and regional coordination teams, which determine how to best support travelers and directly communicate with them, Tatum said. The agency rolled out the TSA Cares app last summer.

With the technologies, Salesforce projects 70,000 internal data-sharing transactions will be automated annually, freeing up over 11,660 staff hours. TSA’s applications are expected to save $11 million over the next three years, according to Salesforce.

“If we can help the traveling public easily access accurate, timely information, we can help alleviate stress in the field on travel days and increase the TSA’s overall effectiveness,” Niki French, customer experience branch manager at TSA’s Office of Civil Rights and Liberties, Ombudsman and Traveler Engagement, said in the blog post.

]]>

Supreme Court keeps USPS from carrying out mail-in voting changes in midterm elections

Jory Heckman — Tue, 15 Sep 2026 17:31:00 -0400

The Supreme Court is keeping the Trump administration from imposing new rules for mail-in ballots in the lead-up to the midterm elections this November.

President Donald Trump signed an executive order in March imposing new rules on mail-in voting. The Postal Service released a final rule in August to implement the executive order.

The court’s tersely worded and unsigned order issued Monday evening states that the government “is unlikely to succeed on the merits of its challenge” to a lower court's preliminary injunction that barred USPS from taking further steps to carry out its final rule.

The final rule would have required states planning to send mail-in or absentee ballots to provide USPS with a list of participating voters. The final rule also established new barcode and envelope-formatting requirements.

USPS previously provided nonbinding guidance to states on envelopes and barcodes in previous elections, but its final rule would have allowed the agency to reject ballots that didn’t meet its standards. USPS was also tasked under the executive order with launching a Federal Mail Ballot Portal for states to upload their voter lists.

Justice Brett Kavanaugh wrote in his concurring opinion that "there is at least a fair prospect that the final rule falls within the Postal Service’s statutory authority," but that applying the final rule to this year’s elections would be “arbitrary and capricious” under the Administrative Procedures Act, “because state and local election officials do not have sufficient time to reasonably implement the rule before the elections.”

Justices Samuel Alito and Clarence Thomas, in their dissent, wrote that the “Postal Service has broad authority to regulate the mail,” and that without a stay of the lower court’s ruling, the government “will suffer irreparable harm." The USPS final rule, they argued, would have increased the visibility of ballots in transit and would have helped “better detect election fraud.”

Postmaster General David Steiner told Government Executive in an exclusive interview last month that USPS “is not a political organization,” and would comply if the courts decided to permanently block the executive order's implementation.

“We deliver the mail. Someone else makes the rules. We deliver the mail, and so we deliver the mail under the rules that are put on us, and that's exactly what we'll do. People can fight over whether the rules are valid or not. Once that fight is finished, we'll do whatever the courts say we can do. That's not our decision,” Steiner said.

The American Postal Workers Union, which represents the frontline USPS employees most directly impacted by the final rule, warned that its employees hadn’t received training on changes outlined in the executive order or USPS final rule.

In an amicus curiae brief, the union warned that the final rule would “likely cause catastrophic wholesale non-delivery of mail to hundreds of thousands of voters,” because postal clerks “cannot realistically be trained" on these changes ahead of the November elections.

APWU represents the frontline Business Mail Entry Unit (BMEU) clerks who accept large batches of mail-in ballots to be delivered to voters. Under the final rule, postal clerks would have scanned a sample of each batch of ballots to confirm the voter information on individual ballots matches the information states submitted to the online portal.

A USPS whistleblower claimed that if clerks identified any errors in the sample, USPS would have refused to accept the entire batch - in many cases, tens of thousands of ballots - until election officials resolved the errors.

APWU also represents the IT workers tasked with creating that online portal. The USPS whistleblower also claimed that the agency has rushed to complete work on the portal, and hasn’t followed “standard principles of testing and debugging new software before launch.”

]]>

OPM to continue ChatGPT deployment following reupped OneGov deal

Christian Robles — Tue, 15 Sep 2026 12:42:00 -0400

The Office of Personnel Management will “continue to lean in” on deploying artificial intelligence — including ChatGPT — following the General Services Administration announcement that a new OneGov partnership with OpenAI will offer agencies reduced consumption-based pricing, instead of the previous $1 a year price, starting on Oct. 1.

“I want to continue to offer everyone at OPM multiple tools so that, as the frontier changes, people can move to the best tool, and then vendors can compete for our business every day by offering the best tool,” said Adam Starr, OPM’s chief information officer, at the GovExec Government & AI Summit Tuesday.

Starr said private companies should work with GSA on inking OneGov agreements, particularly as it makes procuring technology easier for smaller agencies such as OPM.

“I feel like I’m getting the right deal,” he said of the GSA initiative.

GSA is currently moving into the next phase of its OneGov program — designed to cultivate deals on technology by treating the government as a single, large customer — as it looks to extend expiring agreements and expand the initiative beyond software. Deals for Anthropic’s Claude and Google’s Gemini are set to expire at the end of the month.

OneGov critics have argued that, as federal employees become accustomed to low-cost technologies, agencies will be locked into continuing to provide access to the technologies and have to absorb higher costs when OneGov deals expire or deal terms change, critics contend.

“With this shift from fixed cost to a consumption model, we are working on setting appropriate limits on spend. We aim to keep multiple frontier AI tools in the hands of every employee,” an OPM spokesperson told Nextgov/FCW when asked how the new OneGov partnership with OpenAI will affect agency costs.

AI adoption at OPM

In addition to ChatGPT, OPM employees currently have access to three other large language models per the agency's AI inventory: Copilot, Grok and Claud.

More than 70% of employees are “active users” of at least one of those AI tools, according to Starr.

OPM’s June workforce pulse survey found that the agency had a “net promoter score” 0f 17.9 on a scale of -100 to 100, according to the OPM spokesperson, suggesting employees have a favorable view of the AI tools. The score captures user sentiment of AI by asking employees how likely they are to recommend an approved AI tool to a colleague. The Office of the Director had the highest net promoter score of +95.7 while the human capital office was close behind at +64.1, according to the spokesperson.

OPM is using AI for a variety of purposes.

As the agency looks to consolidate 119 human resources IT systems into one, one employee is using AI to distill thousands of pages worth of training materials from across the federal government for the new system, said Starr. OPM employees have used AI to build dashboards and rewrite the Customer Relationship Management platform deployed at retirement services contact centers, he continued.

To support the adoption of AI at OPM, the agency has stood up training initiatives for employees.

Roughly a quarter of the agency has signed up for an ambassador program where employees learn how at least two AI platforms apply to their work and teach five to 10 other people about the technologies, said Starr. The agency also allows vendors to train employees and conduct its own in-house training on topics such as AI platforms and agentic AI.

Underpinning OPM’s AI adoption strategy is a framework known as STEPS, according to Starr.

The first S stands for signals, which refers to the metrics used to guide AI adoption. The T stands for tools, which refers to ensuring agencies have access to cutting-edge technologies. The E stands for enablement, which covers training, ambassador and other programs that help employees make the most out of AI. The P stands for policies, such as OPM’s recent AI hiring memo, that explains what AI can and cannot support. The second S stands for staffing agencies with the right talent through hiring initiatives such as Tech Force.

The STEPS framework has informed an OPM program called the AI Governance and Adoption program, or AI GAP, which seeks to help other agencies harness AI, said Starr.

Editor's note: This article has been updated to note the number of large language models mentioned in OPM's most recent AI use case inventory.

]]>

New bipartisan Senate science caucus will address AI

Alexandra Kelley — Mon, 14 Sep 2026 19:01:00 -0400

On Monday, Sens. John Hickenlooper, D-Co., and Dave McCormick, R-Pa., jointly unveiled a new Senate Bipartisan Science and Innovation Caucus that aims to investigate and support U.S. scientific research and innovation – and weigh in on AI policy.

While the caucus’ focus will hit a wide array of topics, Hickenlooper said that he anticipates the caucus examining AI policy issues like regulation.

Speaking to Nextgov/FCW at the caucus launch event, Hickenlooper acknowledged the pressing global rivalry with countries like China to develop the most advanced AI systems, but he said it doesn’t negate AI’s threat potential.

“We want to be careful and make sure that we don't have something running amok that could cause large amounts of damage,” Hickenlooper said, adding that the new caucus will start working on these issues soon to keep up with AI developments.

As the ranking member on the Senate Commerce Subcommittee on Consumer Protection, Technology and Data Privacy, Hickenlooper acknowledged that there is overlap in committee jurisdiction over different AI subject matter.

In addition to AI, the caucus will focus on promoting policy that advance aspects of the U.S. scientific enterprise; examining its potential benefits and threats; its economic and national security impact; and how legislation could benefit it.

“I think you look at the things that distinguish America,” McCormick said. “Most of them are investments in science and a parallel expansion into the growth of entrepreneurship at the highest level.”

The Science and Innovation Caucus follows news that House Democrats are contemplating the creation of a new select committee on AI if they take control of the chamber following midterm elections.

]]>

NSA plans major reorganization with new AI, China and cyber centers

David DiMolfetta — Mon, 14 Sep 2026 18:11:00 -0400

The National Security Agency is preparing a sweeping reorganization that would establish five mission centers focused on artificial intelligence, China, cybersecurity, combat support and global intelligence, according to a person familiar with the matter.

The plan, developed under NSA head Gen. Joshua Rudd, would put new mission directors in charge of those areas as the agency adjusts its structure around emerging technology and national security priorities.

The person spoke on the condition of anonymity to communicate details of the new arrangement. It would mark one of the most significant organizational changes at the intelligence agency in a decade.

The Washington Post first reported the planned overhaul Sunday.

“While detailed personnel specifics cannot be provided at this time, NSA’s focus is to ensure that the organizational restructuring of Agency priorities enhances the Agency’s ability to fulfill its critical national security missions at both speed and scale,” an agency spokesperson said, noting that the agency’s “core missions remain unchanged.”

New mission directors are expected to submit organizational plans in the coming weeks, with implementation of the structure beginning in October and a goal of reaching full operational capacity some time in January of next year.

Some key details appear to remain unsettled, including whether the centers would replace existing divisions or add a management layer over the agency’s current structure.

The agency collects and analyzes intercepted foreign communications and other electronic signals to inform U.S. national security decisions. It also helps protect sensitive government networks and encryption systems against foreign threats.

Rudd, who was confirmed in March, also leads U.S. Cyber Command, the Defense Department unit responsible for offensive and defensive cyber operations. It’s not clear how the reorg would affect operational relationships between the combatant command and the foreign eavesdropping giant. Both organizations are headquartered at the sprawling Fort Meade campus in Maryland, with personnel extending across the United States and overseas.

Under the plan, the NSA’s elite Tailored Access Operations hacking unit is expected to be lodged under the global intelligence arm and receive additional resources once the federal government’s budget resets at the end of this month. The unit, recently reconstituted under Rudd, specializes in penetrating foreign digital targets.

The proposed AI center comes as NSA officials publicly describe a growing role for the technology in both intelligence analysis and cybersecurity.

At the Billington Cybersecurity Summit last week, cybersecurity directorate chief David Imbordino said the agency is exploring AI tools to help analysts sort through enormous quantities of collected information and identify useful material faster, and he said the technology could accelerate analysis that currently takes days or weeks.

NSA Deputy Director Tim Kosiba similarly said last month that the agency is engaging leading commercial developers as it seeks broader access to their systems. New AI models are increasingly capable of finding software flaws and writing code to exploit them, prompting global debates over safety standards, who should have access to the tools and how governments should use them.

“We want access to all the models, and we’re going to take advantage of that,” Kosiba said at an intelligence conference in Bethesda, Maryland.

NSA is also helping implement a June executive order establishing a voluntary process for government scrutiny of advanced AI systems’ hacking capabilities before their wider release. The agency’s director is responsible for determining which models qualify for that process, in consultation with other federal offices.

The restructuring follows an earlier overhaul, known as NSA 21, launched in 2016 under then-Director Mike Rogers. That effort sought to reorganize the agency into six directorates.

]]>

Agriculture CIO transitions to role modernizing farmers’ records

Edward Graham — Mon, 14 Sep 2026 18:08:00 -0400

The Agriculture Department’s top IT official is stepping down from his position to join an agency modernization initiative, a USDA spokesperson confirmed to Nextgov/FCW on Monday.

Chief Information Officer Sam Berry officially joined USDA last September, when he replaced longtime agency CIO Gary Washington. Berry previously served as an advisor at Agriculture for the cost-cutting — and controversial — Department of Government Efficiency.

“Sam Berry has been a tremendous USDA CIO,” an agency spokesperson said in a statement. “He will now be assuming a new role at USDA where he will be focused on the implementation of One Farmer One File.”

USDA announced the technology modernization initiative in February, saying at the time that the effort is designed “to create a single, streamlined record that follows the farmer — no matter where they go in the USDA system.”

Agriculture subsequently announced a $300 million agreement with Palantir in April to, in part, support the rollout of the One Farmer, One File initiative.

“Protecting America’s farmland is protecting America itself, and this work gives USDA the visibility and speed needed to safeguard our food supply,” Berry said in a statement about the partnership at the time. “Our farmers sustain this nation, and modern tools help us support them with greater precision. I look forward to working with Palantir as we continue serving the American farming community, which serves all of us every single day.”

Federal News Network reported on Sept. 8 that USDA chief information security officer Tony Brannum will serve as interim CIO. Agriculture did not disclose the nature of Berry’s new position with the One Farmer, One File initiative. The agency has reportedly been reaching out to other internal tech and IT officials in recent weeks about taking on roles with the initiative.

]]>

Trump redoubles pro-AI stance amid calls to slow development

Alexandra Kelley — Mon, 14 Sep 2026 17:16:00 -0400

President Donald Trump on Monday pushed back against major tech leaders’ recent calls to slow the pace of artificial intelligence development for the sake of safety, saying the U.S. does need “strong and smart” guardrails but that the country is already outfitted with the proper safety infrastructure.

In a Truth Social post, Trump wrote that the “only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!” He added that his administration has been working to stop AI developers from “doing bad, or potentially bad ‘things’” and that current U.S. regulatory structures already offer a level of oversight that can ensure AI won’t be used to cause harm. Trump also doubled down on the national security impetus to ensure the U.S. is dominant in the creation of new AI products over other countries like China.

Trump’s comments follow public posts from both Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman concluding that the broader AI industry needs to slow its research into potentially dangerous AI products.

Amodei first wrote in a weekend blog post that AI developers should “pace the frontier” in order to ensure that the technology’s innovation doesn’t outpace the ability to deal with threats it may pose.

“The most effective method of pacing is via regulation that targets all US frontier AI companies, as that covers even those who are unwilling to cooperate voluntarily,” Amodei wrote.

OpenAI CEO Sam Altman similarly warned against the ways “AI progress could go very badly,” in an early Monday X post, saying that safety techniques need to stay ahead of progress on model capabilities and that concentration of power could cause “dystopian” results.

“We already have tremendous CRIMINAL and REGULATORY power over these companies!,” Trump wrote. “There is a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China. WHOEVER WINS AI, WINS! We are leading China, and all others, and will continue to do so. Conspiracy Theorists, Treasonists, Traitors, and Leakers, BEWARE!”

Trump followed this initial post with more commentary, calling the warnings of AI threats — which were catalyzed by a series of Tuesday X posts made by former OpenAI and Anthropic employee Jacob Coxon — part of a larger “hoax.”

“Concerning AI, when, in the History of Business, did anyone see the Leaders of an Industry call for Regulation that, if strongly implemented, will drive them into oblivion and bankruptcy?” Trump wrote on Monday. “AI taking over the World, destroying Humanity, and all other things bad, is a HOAX.”

Coxon’s posts following his resignation from Anthropic over safety concerns went viral and have captured Congressional attention, with multiple lawmakers calling for the prioritization of AI regulatory legislation on X.

As Trump prepares to welcome Xi Jinping, the president of the People’s Republic of China, to the White House on Sept. 24, he wrote that China isn’t slowing down their development of new AI technologies, and he plans on keeping with his pro-AI agenda to ensure U.S. dominance in the market.

Speaking on CBS Sunday Morning, Amodei, who has long been an advocate for responsible AI development, said that the burden of ensuring safe AI is not confined to just the AI industry or the federal government.

“I feel both sides of the responsibility,” Amodei said. “It shouldn’t be my responsibility alone or Anthropic’s responsibility alone or all of the AI company’s responsibility alone or even one government’s responsibility alone. This is bigger than all of us.”

Trump, however, made clear his regulatory posture would continue to favor industry growth.

“The only reason the AI/Data Center outburst is happening is because the United States is leading, by a lot, every other country,” Trump said on Truth Social. “Don’t kill the Golden Goose!”

The White House did not respond to a request for comment by the time of publication.

Industry experts are also divided on the path ahead for AI regulation. The Center for Democracy and Technology echoed Amodei and Altman’s call to pace the frontier of AI tech development and supported more federal regulatory mechanisms.

“Companies clearly need stronger incentives to ensure AI systems won’t cause harm to individuals, critical infrastructure, and society more broadly,” CDT President & CEO Alexandra Reeve Givens said in a statement. “Especially concerning is the White House’s approach of secret, arbitrary processes and vague threats of retaliation, which is not only ineffective but also undercuts the cornerstones of democracy. Governance mechanisms must be effective, legitimate, and collaboratively designed, providing transparency and a foundation for public accountability.”

Adam Thierer, a senior policy analyst at the R Street Institute, sided with the Trump administration, writing on X that surrendering the competitive lead the U.S. has in AI technology is “incredibly dangerous thinking,” particularly if its development is dictated by a “combination of governments,” for diversified oversight, as Amodei suggested on CBS Sunday Morning.

]]>

VA watchdog uses AI to help compile healthcare facility audits

Edward Graham — Mon, 14 Sep 2026 15:00:00 -0400

The Department of Veterans Affairs’ watchdog has been using Microsoft Copilot Chat to assist in its audits of VA healthcare sites, with the artificial intelligence tool being leveraged by investigators to review documents, transcripts and other collected data.

The AI-powered assistant has been deployed by the VA Office of Inspector General’s Office of Healthcare Inspections, which evaluates VA medical facilities to ensure veterans are receiving proper medical care and mental health services.

Since March, OIG’s hotline, healthcare facility and mental health inspection reports have included the same boilerplate language disclosing the use of the AI tool, which states that, “during the preparation of this report, the inspection team used peer-reviewed standardized, structured, and evaluated prompts in Copilot Chat (Microsoft) to review inspection data such as interview transcripts, documents, questionnaire responses, and physical observations.”

The standard wording goes on to say that, following use of the tool, watchdog investigators “confirmed fidelity of the generated output to the source material, edited the report, and took full responsibility for the content of the publication,” and that “all references are for original source material, not AI-generated content.”

An April 2025 memo released by the Office of Management and Budget called for agencies to take steps to manage high-impact AI uses, which it defined as instances “when its output serves as a principal basis for decisions or actions that have a legal, material, binding, or significant effect on rights or safety.”

Because VA OIG’s Office of Healthcare Inspections teams “do not use AI as the principal basis for decision-making or actions,” however, the report says the use of Copilot Chat “does not meet the definition of high-impact” outlined in the OMB memo.

A spokesperson for the watchdog said the AI tool only helps the watchdog to more efficiently review the vast amount of data and information that it collects during the course of its healthcare site investigations. They added that “data entered into this tool is not shared with the Department, nor does it leave the VA OIG secure environment.”

In an interview with Nextgov/FCW, former VA OIG Inspector General Michael Missal said the watchdog tested Copilot Chat and purchased it under his watch, with investigators ultimately beginning to use it in 2024.

“We knew VA had used it in some capacity, and what we were trying to do was to use it to make our work more efficient and make it go more quickly,” Missal said about the tool.

Prior to being fired by President Donald Trump in January 2025 — along with more than a dozen other federal agency IGs as part of a controversial mass purge — Missal said his goal was to ultimately expand Copilot Chat’s use to the OIG’s Office of Audits and Evaluations and Office of Investigations.

A review of other OIG-published reports over the past year that were conducted outside the purview of the Office of Healthcare Inspections does not appear to show any other audits disclosing the use of the AI tool. But Missal said Copilot Chat is uniquely suited for use by the watchdog’s Office of Healthcare Inspections, which is why that unit was the first to adopt it.

He noted that OIG personnel conducting site audits of VA medical facilities often investigate the same issues or look into the same on-site areas, so creating standardized prompts for the tool helped provide the office with consistency.

“It's a time-saving way [to conduct inspections], and it's also quality assurance as well because this way everybody's operating with the same prompts, and it just ensures you're not missing something,” he said. “What this does is it prompts you: ‘What about this? What about this? You get this. What about that?’ You know, [it] just takes you through a whole menu of things.”

Both the OIG and Missal said Copilot Chat’s output does not make any final determinations or write any of the final report.

“VA OIG requires human review of AI output to ensure accuracy prior to the inclusion in any published report,” the watchdog’s spokesperson added. “Our inspection and audit teams thoroughly review AI output and confirm accuracy.”

Missal also said investigators using Copilot Chat “check everything with the source, so that you know whatever we put in a report, we made sure it was not AI generated, but actually from the work.”

“It was just easier to see trends and to look at big pictures by using this program, but then ultimately, in what we relied on in the report, we would make sure we would source the actual interview document,” he added.

The watchdog’s 2025 AI use case inventory lists 20 different examples of the emerging capabilities, with one instance — using AI-assisted tools in word processors — being identified as leveraging Copilot Chat.

OIG’s AI strategy, which was similarly released by the watchdog in September 2025 to comply with OMB’s memo, says that it “utilizes AI in meaningful ways that enhance the office’s mission of conducting independent oversight that improves the health and welfare of veterans and their families, caregivers, and survivors by strengthening VA programs and operations.”

As Missal noted, VA has also made Copilot Chat available to its staff, including those within its healthcare unit, the Veterans Health Administration. A June OIG report warned that VHA personnel were using the tool for clinical use, despite the agency not listing Copilot Chat as a high-impact AI use case.

]]>

The internet of (compromised) things: securing IoT when you can’t trust (any) “thing”

Don Parente — Mon, 14 Sep 2026 14:35:00 -0400

A few years ago, I was at an industry conference attending a cybersecurity session when the presenter referred to the Internet of Things as the Internet of (Compromised) Things.

The audience laughed. But, like all good humor, it was funny because it was true.

The Internet of Things has transformed the way we monitor, manage and operate everything from factories and utilities to buildings, vehicles and supply chains. Sensors, cameras, meters, beacons, probes and controllers are now connected to networks and continuously sending data back to applications that make our businesses smarter and more efficient.

But there is a fundamental cybersecurity problem hiding inside all of that connectivity: we have dramatically increased the number of devices connected to our networks without necessarily increasing our ability to secure them.

Over the years, we have seen numerous vulnerabilities in the software embedded in IoT and industrial devices. One particularly striking example was Ripple20, a collection of vulnerabilities discovered in the Treck TCP/IP stack used in a wide range of IoT and industrial products. Some of the vulnerabilities could be exploited remotely, including one with a CVSS severity score of 9.8. CISA’s recommended mitigations included minimizing network exposure, putting control-system devices behind firewalls and isolating them from business networks.

And that brings us to an uncomfortable question. Is industrial IoT really more secure than the IoT devices we have in our homes?

It is tempting to assume that the answer is yes. After all, commercial and industrial solutions are generally designed for professional environments, purchased through established vendors and deployed with more formal IT and operational processes. But I don’t think that is a safe assumption.

In fact, industrial IoT can present a particularly difficult cybersecurity challenge. Unlike our PCs and smartphones, which routinely tell us when an operating-system or application update is available, many IoT endpoints are designed to perform a very specific function and then remain in service for years — sometimes decades.

Consider:

A sensor doesn’t necessarily have an IT department.
A controller doesn’t necessarily have an automatic patch-management system.
Taking an industrial device offline to install a firmware update may not be as simple as clicking “Restart Now.”

The result is that vulnerability identification, remediation and patching may happen far less frequently — or, in some cases, not at all. So perhaps we should approach the problem differently.

What if we assume the IoT endpoint will eventually be compromised? That may sound pessimistic, but it leads to a much more realistic and resilient security architecture. Rather than relying solely on the device itself to remain secure, we can use the network surrounding the device to contain the risk.

I would suggest three network-centric principles for securing IoT endpoints:

Keep the “things” away from the people.
Control the flow of traffic between the “things.”
Monitor the network behavior of the “things” and mitigate when necessary.

This is not a foolproof approach. As we like to say in cybersecurity, nothing connected to the Internet can ever be made 100% secure. However, we can make it considerably harder for a compromised IoT device to become a compromised network.

1. Keep the “things” away from the people

At this point, it is fairly well-documented that human error is a major contributor to cybersecurity incidents. So why would we put our IoT devices on the same network as the people?

An IoT network should not be allowed to freely co-mingle with the broader administrative network that supports email, financial applications, collaboration tools and other systems used by employees.

Instead, the “things” should live in their own network environment, separated from the people and the applications they use. That could mean a dedicated VLAN, a separate VRF or, more broadly, a dedicated VPN and security policy for IoT traffic.

For cellular-connected devices, technologies such as Private APNs — Access Point Names — can provide another layer of separation. A Private APN allows traffic from cellular-connected devices to flow into a private network rather than traversing the public Internet.

The objective is simple: If an employee falls for a phishing attack, don’t let that mistake take down your IoT network.

Network segmentation creates a barrier between the administrative environment and the IoT environment, limiting the ability of a compromised user device to reach thousands of connected endpoints.

2. Control the flow of traffic between the “things”

Most IoT applications are fundamentally hub-and-spoke architectures. Sensors, beacons, probes, meters and other endpoints collect information and report it back to a centralized server or cloud application. The endpoint generally doesn’t need to communicate with every other endpoint on the network.

Traffic between IoT devices should be restricted wherever possible. A temperature sensor should be able to send temperature data to the application server. It probably doesn’t need to initiate a connection to another temperature sensor across the building.

The same principle applies to cameras, meters, controllers and other connected devices. By restricting east-west traffic between IoT endpoints, we can limit the ability of one compromised “thing” to compromise another.

This is particularly important because once an attacker gains control of an endpoint, the attacker may use that device as a launching point to discover and attack other devices.

The goal should be minimum access privilege for machines, just as we apply minimum access privilege to people. A thing should be allowed to communicate with the things and applications it needs to perform its job — and nothing more.

3. Monitor the network behavior of the IoT “things” and mitigate when necessary

Even with segmentation and traffic controls, we should operate under the assumption that something will eventually get through. That makes behavioral monitoring the third layer of defense.

Modern cybersecurity tools are capable of modeling expected network behavior and identifying activity that falls outside the normal pattern. IoT devices are particularly well suited to this approach because their behavior is often highly predictable.

Consider a sensor that normally communicates with an applications server once every hour. That is its normal behavior. If that same sensor suddenly starts communicating continuously, something has changed. If a device that normally sends a few kilobytes of telemetry suddenly begins transmitting hundreds of megabytes, something has changed.

And if a sensor whose only job is to send data suddenly attempts to log into another server, scan the network or communicate with dozens of other devices, something has definitely changed.

Those behavioral changes should generate an alert — and, where appropriate, trigger an automated response. The network should be capable of saying, in effect: “That’s not what you normally do, so I’m going to stop you.”

That could mean blocking the offending traffic, quarantining the device, restricting it to a limited set of destinations or otherwise containing the endpoint until it can be investigated.

Assume the thing will be compromised

The traditional approach to IoT security has often focused on making the device itself more secure: better passwords, stronger authentication, secure firmware and regular patches. All of those elements are important. But they aren’t enough.

We need to recognize the reality of IoT deployments. Some devices will have vulnerabilities that aren’t known when they are deployed. Some vulnerabilities won’t be patched quickly. Some devices won’t be patched at all. And some devices will remain in service long after their original manufacturer has stopped supporting them.

The answer, therefore, isn’t simply to build a better “thing.” It is to build a network that assumes the thing might not be trustworthy.

The Internet of Things isn’t going away. In fact, it is becoming an increasingly important part of how we operate our businesses and critical infrastructure.

The challenge for cybersecurity professionals is to make sure that the convenience and intelligence created by billions of connected devices don't come at the expense of the security of everything else.

Maybe the presenter at that conference was right. Perhaps we really are building an Internet of (Compromised) Things. But with the right network architecture, a compromised thing doesn’t have to become a compromised enterprise.

]]>

VA to gradually expand medical AI scribe following successful pilots

Christian Robles — Mon, 14 Sep 2026 13:59:00 -0400

The Department of Veterans Affairs is planning to expand the use of an artificial intelligence tool that records veteran-doctor conversations with patient consent after a series of rollouts it says have been successful.

The VA first deployed the technology — known as ambient AI scribes — in October 2025 at 10 medical centers across the nation. After a series of pilots over the past year, the tools are now available to all VA Patient Aligned Care Teams’ primary care providers, behavioral health interdisciplinary program practitioners, physical medicine rehabilitation practitioners and medical and surgical specialists.

“Each time they do a pilot with some volunteers to make sure it's working well for that particular discipline, and then after that they'll do some specialty testing on certain features with an early adopter group,” Dr. Rebecca Gladding, acting deputy chief of psychiatry at the VA Greater Los Angeles Healthcare System, told Nextgov/FCW.

The department plans to bring ambient scribes to select outpatient care providers and broaden its use through an enterprise contract this year. Dr. Gladding told Nextgov/FCW that expanding the technology to inpatient settings is not currently part of department plans because “it’s a lot more complicated, a lot more data” than outpatient settings.

When asked how the VA decides to scale the use of ambient scribes, a department spokesperson said it “evaluates candidate specialties against both the data and the clinical fit with the goal of relieving administrative burden for as many clinicians as possible and giving them more time to engage with veterans.”

VA providers are currently using ambient scribes from Abridge and Knowtex to record conversations with veterans, allowing them to look at their patients rather than their computer screens. The recordings — which veterans must verbally consent to after an explanation from their providers — are turned into electronic health record notes. Veterans can opt out of the technology at any point, including mid-visit.

Abridge is the only vendor whose ambient scribe is live at VA medical centers, operating on both legacy VistA and new Oracle-based EHRs, according to vice president of partner experience Reba Schenk. The company deferred to VA when asked about uptake and usage of its ambient scribe but noted its technology is deployed at 70 VA sites and over 300 U.S. health systems.

Knowtex’s technology has been used in 10 Veterans Integrated Service Networks, or VISNs, and 79 VA medical centers in the past six months, reaching more than 7,000 clinician users and saving over 450,000 hours of EHR documentation time, the company told Nextgov/FCW.

The company has a user adoption rate of 88% with an average satisfaction rating of 4.48 out of five across over 5,100 ratings, which — along with a Kansas City pilot — shows a “successful partnership and implementation at national scale,” the company added.

Pilot results

Pilot data show veterans report better care when a provider uses an ambient scribe, and clinicians report feeling less burnt out by EHR documentation needs.

In the three-month Kansas City pilot, for example, veterans reported a 95.8% satisfaction rate, which is almost 3 points higher than before the evaluation period. Survey results also showed about 95.4% of veterans felt providers’ explanations were clear and 95% of veterans felt they could trust their primary care team. Those numbers are about 2.3 points and 2.9 points higher than before the evaluation period, respectively.

Over 986,000 VA primary care appointments, including in-person, phone and video visits, have used ambient scribes with vendors reporting veterans have declined the use of the technology less than 1% of the time since October 2025, according to Dr. Gladding.

In early ambient scribe deployments, VA clinicians reported lower cognitive load, reduced after-hours documentation burden, improved focus during encounters, less mental fatigue with documentation demands and streamlined note documentation workflows, according to a VA spokesperson.

The VA did not provide data on the quality of ambient scribe-generated notes, though a recent Veterans Health Administration-funded study found AI-generated notes had lower quality scores than human-generated ones across five standardized care cases.

Getting quality notes is “where the learning curve is when you first start using it,” Dr. Gladding said, noting there are several quality checks in place for ambient scribes.

She said clinicians review and approve ambient scribe notes before they are added to EHRs. During the review, clinicians can check conversation transcripts to ensure they are accurately captured by notes. Dr. Gladding added that VA providers have internal conversations around prompt engineering to ensure ambient scribe notes are to clinicians’ liking.

Before providers deploy ambient scribes, they must complete a “one-time prerequisite” ethics training and a technology setup training within VA’s Talent Management System, according to a VA spokesperson. They may also choose to take vendor-specific training, they added.

For ambient scribes, “we have a lot of change management and literacy and training around for our healthcare providers on how to use that tool specifically and how to get the most value out of it,” said Dr. Kimberly McManus, VA’s acting chief AI officer, at the GovCIO Media and Research Health IT Summit earlier this month.

]]>

Tech bills of the week: Community protection from data centers; AI labs for education; and more

Alexandra Kelley and Edward Graham — Fri, 11 Sep 2026 17:43:00 -0400

Guiding data center development

A new legislative package looks to address government management of data center construction amid artificial intelligence-motivated demands for increased compute capacity. The National Data Center Plan, introduced on Tuesday by Rep. Suhas Subramanyam, D-Va., is composed of four distinct bills that address different concerns surrounding data center construction.

The first, the Responsible Data Center Siting Act, directs the Energy secretary to create best practices for choosing sites for new data centers, taking issues like energy availability, noise pollution, workforce needs and water and electricity use into account.

The second, the Data Infrastructure Energy Measurement and Standards Act, also focuses on resource usage and asks the National Institute of Standards and Technology to develop technical standards for accurately measuring data center energy and water use.

The third, the Data Infrastructure Risk Reduction Act, requires leadership from both the Department of Homeland Security and the Cybersecurity and Infrastructure Security Agency to collaborate with the Defense secretary in developing a defense strategy to protect communities living near data centers. The strategy will also address how to protect the U.S. data infrastructures from adversaries.

The final bill, the Data Center Fair Share Act, requires data centers and large energy consumers to pay the full cost of grid upgrades needed to handle growing compute loads. The bill would call for states to adopt large-load tariffs to ensure that data centers pay for the amount of power they consume.

“People are concerned about the future of our community and angry about the impact this will have on noise, the environment, and their property,” Subramanyam said in a press release. “The approach to building data centers has been driven by short-term profit rather than long term planning. We need a national plan that puts communities first rather than the current ad hoc approach.”

Creating AI labs at K-12 schools

Rep. Josh Gottheimer, D-N.J., officially introduced a measure on Tuesday that seeks to create dedicated artificial intelligence labs at K-12 schools to help students and educators better use and understand the emerging capabilities.

The bill calls for the Education Department to establish a new federal grant program — or direct resources from other existing grants — to help develop the labs and study how the provided funds are used to guide future education-focused AI investments.

Gottheimer’s proposal was introduced amid the launch of his “Back-to-School Agenda on AI and Screen Time in K-12 Education,” which he said is designed to enhance the next generation’s technology literacy.

“AI can be a powerful tool, but it can never replace the fundamentals,” he said in a statement. “Give our schools the tools. Give our teachers the training. And give our children a real head start, instead of a false choice between banning technology outright and drowning in it.”

Preparing students for a data-driven future

A bipartisan quartet of House lawmakers rolled out a measure on Thursday that seeks to bolster data science educational opportunities for educators and students from pre-K to community college.

The Data Science and Literacy Act — introduced by Rep. Haley Stevens, D-Mich., and co-sponsored by Reps. Jim Baird, R-Ind., Don Beyer, D-Va., and Brian Fitzpatrick, R-Pa. — would establish a grant program within the Education Department to fund the creation of data science-focused curricula, student support opportunities, professional development training and education partnership opportunities.

“Data is powering the technologies and jobs that will define America’s economic future, and our students deserve the skills to lead that future,” Stevens said in a statement. “Whether a student goes on to work on the factory floor, in a research lab, at a small business, or developing the next breakthrough technology, the ability to understand and use data will be increasingly important.”

]]>

‘Digital employees’ are coming to the Defense Logistics Agency

Frank Konkel — Fri, 11 Sep 2026 17:40:00 -0400

The Defense Logistics Agency is piloting AI agents that work autonomously—bots that its CIO calls “digital employees.”

“This is a new frontier for everyone,” DLA chief information officer Adarryl Roberts told reporters at DLA’s Industry Collider Day on Sept. 9 in Alexandria, Va.

The move builds on DLA’s increasing use of robotic process automation, or RPA, and the generative AI tools available to its 25,000 or so military and civilian personnel. In 2025, bots saved DLA an estimated 300,000 hours of work—some working around the clock to help the combat support agency get everything from food and fuel to medical supplies and clothing where it needs to go.

“We have approximately 185 to 190 bots that are running. I’d say about 90% to 95% of those are unattended bots,” Roberts said. “So, we really cracked the nut to have those unattended. Now, we’re moving from the bots to, ‘How can we have agentic agents working across [the department], based upon our data? So, we’re early in that journey.”

Military-specific versions of ChatGPT and Grok recently joined Gemini on the platform that gives defense workers access to generative-AI tools for information that is sensitive but unclassified. DLA gets its workers up to speed on these tools through “GenAI 101 training,” Roberts said.

“I would look at generative AI as the concept of making the workforce and agency—from an operating model—comfortable with the results you’ll get with AI,” he said. Tools like Gemini and ChatGPT “allow the workforce to interact with AI, see the results, trust the results, and get comfortable.”

“That’s required, because as you move to a more agentic AI, those agents are working 24/7 behind the scenes, and so there has to be a level—a cultural level—of trust in those outcomes,” he said. “Now, we’re shifting to what I would call strategic implementation in terms of those agentic agents to get us better results.”

Roberts said introducing digital employees will require new management strategies. Humans will be working alongside—and often overseeing—digital agents, creating an assortment of new management challenges.

“I think we like to call it, ‘How are we managing our digital employees?” Roberts posited. “Because that’s what they really are, and how do you manage them in conjunction with your human employees?” How do you train a workforce to where everyone is essentially working as a supervisor because of the employment of AI with humans? Do you terminate an agent because of the performance?”

“So, it's a very complex universe, from an employee-cultural perspective, in terms of how you deploy and manage those agents,” Roberts said.

It might sound like science fiction, but the rapid evolution of AI—and DLA’s gradual adoption of it—creates questions around governance, rules, security, and the performance of AI agents and even human employees.

Many of these questions fall to DLA’s AI Center of Excellence, founded in 2024 to ensure the safe, secure, and responsible integration of AI across the agency. The center oversees DLA’s use of AI and performs other functions, like ensuring agents are being trained properly. DLA has implemented a zero-trust cybersecurity framework that Roberts called “foundational” to high-level agentic work.

Currently, the agency is establishing “persona-based access” that will dictate AI agents’ account permissions based on their job descriptions. For example, a procurement specialist at DLA may require access to certain data domains, like procurement and financial data, so a digital agent performing the same procurement specialist duties should have permission to access only those same data domains.

“These AI agents should have a persona, just like a person,” Roberts said.

DLA is also working to clean up its datasets.

“How do we create a single source of truth for DLA and our stakeholders in terms of our data, making sure it’s a gold standard, meaning we remove the duplicity, errors, etc…so the data is in a usable state, not just for analytics and AI, but as we look toward future system sustainability,” Roberts said.

‘AI is the new gunpowder’

DLA accelerated its use of AI during the 32-month tenure of Director Lt. Gen. Mark Simerly, who will retire on Oct. 2. Simerly will be replaced by a familiar face at DLA, Air Force Lt. Gen. David Sanford, who spent the past two years as director of Logistics Operations and commander, Joint Regional Combat Support.

Speaking at the Sept. 9 event, Simerly said technology allows DLA to be predictive and proactive in solving problems, and he views AI as the key ingredient to solving them at scale.

“I believe when it comes to military logistics, AI is the new gunpowder,” he told reporters. “And we don’t think it’s just an explosive, destructive capability, although we know it can be. It can also be very productive.”

Simerly’s philosophy at DLA has been “to move at pace with the commercial world,” adopting and adapting capabilities rapidly, establish the proper network and infrastructure to access and manage agency data suitably, “and then allow interoperability between the services and with industry to allow us to truly operate end-to-end supply chain operations that are optimized.”

A legacy of change

Simerly said changes at his agency were spurred by the war in Ukraine, which expanded in February 2022, and even during the seven-month-old war on Iran launched by President Donald Trump.

“I think probably the greatest change that I've seen in the agency is a willingness to change within the agency. This recognition that we have to respond to the changing environment, that we have to respond to the changing technology, and that we've got a responsibility to change as rapidly as we can, so we can continue to be relevant and offer support to the warfighters,” he said. “I would say that we were built to change from the very beginning. We've had to accelerate that change cycle in the last few years, and I've seen the willingness and the readiness of the agency to take that challenge on and to operationalize our transformation planning into tangible outcomes.”

The contested logistics in Ukraine captured his attention because they were so different from previous U.S. military conflicts in Iraq and Afghanistan, where DLA delivered food and fuel to forward operating bases “really without great resistance.”

“That would never take place in a place like Ukraine, in the way that they’re operating, the way they have to disperse, the way that the lines of communications are contested and where everything has to be done in a certain way,” Simerly said.

“What sort of sparked some of the transformation energy was very much awareness of what was taking place and being illustrated on that battlefield in Ukraine about the things that we no longer could rely upon, like the air and domain dominance that we possessed in previous fights that is now no longer valid or in place.”

Simerly said similar conditions pertain in the U.S. war on Iran.

“There's no question that we're learning the same lessons now in Operation Epic Fury with Iran and their ability to target a broader array of things with precision that puts much more at risk, and you know certainly we're learning a lot about the execution of alternate supply chains, alternate supply means, and the necessity of being able to operate with precision,” he said.

Simerly expressed pride in his DLA team.

“I think about throughout my career the winning teams that I've been privileged to be part of, and that's what I'm proud of here,” he said. “The winning team that I'm part of now, that will continue on the mission after I leave.”

]]>

Lawmakers clash on safety testing language for in-development AI legislation, people familiar say

Alexandra Kelley — Fri, 11 Sep 2026 16:49:00 -0400

An in-the-works artificial intelligence regulatory bill from members of the Senate Commerce Committee — sponsored by Amy Klobuchar, D-Minn., and John Thune, R-S.D. with input from Chair Ted Cruz, R-Texas — is facing pushback from the panel’s ranking member, Sen. Maria Cantwell, D-Wash., according to multiple people familiar, due to disagreements over the proposal’s AI model safety testing framework.

Multiple people familiar with the bill’s negotiations told Nextgov/FCW that Cantwell’s longstanding position supporting strong federal AI safety standards clashes with provisions laid out in the measure that is under development.

“[Cantwell’s] team has been working with the goal of strengthening the proposal’s testing regime,” one of the people familiar with the proceedings told Nextgov/FCW.

A Democratic committee aide said that the elements in the draft bill that Cantwell wants reworked hinge on having safety tests of AI models be conducted by the companies themselves, who would later present results to the Commerce secretary for deployment approval. Cantwell instead supports language that mandates models instead go through federal agencies — such as the national labs and national security agencies — for security vetting and testing.

“Cantwell really wanted to get these large language models tested and to have mandatory vetting, but the Cruz-Klobuchar-Thune version now that she is opposing, it's primarily a voluntary standard type situation,” the aide told Nextgov/FCW. “Her position all along has been: we need to address catastrophic risk, but we need to do this in a serious way.”

The aide said Cantwell’s office offered language to Klobuchar’s office and described Klobuchar’s office as “very receptive,” but that Cruz’s approval will be a deciding factor.

Klobuchar said in a statement to Nextgov/FCW that AI implementation with little to no oversight comes with “significant risk” and that Congress must act.

“I’m continuing to work toward a bipartisan agreement on legislation for government oversight of the greatest risks posed by AI models,” Klobuchar said. “That includes requiring developers to work with government experts to verify and test models to make sure AI is safe. We cannot allow the release of dangerous models, including those that evade the control of their developers.”

A second person familiar with the matter said Cantwell — along with Anthropic and AI safety groups — are “refusing to play ball,” and that the bill’s provisions already include AI safeguards focusing on catastrophic harm and safety evaluations.

Anthropic did not return a request for comment by the time of publication.

“Republicans have created legal duties for companies to manage catastrophic risks and ensure that government can verify they are doing so and doing so in a way that actually increases safety,” the second person told Nextgov/FCW regarding the legislation. “They are unwilling to create a bipartisan product to give the federal government the power it needs while ensuring continued innovation and due process in the courts. Senator Cantwell is going as far as to require mandatory pre-deployment testing/fixes before down-dais Democrats can support.”

One AI safety group said that their priorities are Congress championing state efforts to promote guardrails in the face of federal preemption.

“We need pre-deployment testing of frontier models and strong national standards that keep pace with fast-growing capabilities and risks,” the safety group told Nextgov/FCW.

While Cruz’s bill with Thune and Klobuchar has yet to be released, Cruz spoke on Wednesday about how AI frontier labs, like OpenAI and Anthropic, are developing their tools and said the alleged lack of safety that a former OpenAI and Anthropic employee detailed in an X post Sept. 8 is “highly concerning,” and that the forthcoming legislation with Thune and Klobuchar will address “catastrophic risk” associated with AI.

“This is scary stuff, but we’re also not going to be able to stick our head in the sand and pretend technology isn’t happening,” Cruz said on Wednesday. “So we’ve got to put some guardrails on it.”

Scores of lawmakers have issued calls for the federal government to effectively regulate the emerging technology, following the former employee’s allegations.

Cruz noted that, despite a need for safety parameters, it is still critical that the U.S. lead the world in AI development, saying that the nation needs to beat China in developing and deploying the latest advanced AI.

Cantwell took to X on Thursday to reiterate her opposition to a “weak federal standard” that undermines existing state AI legislation.

“Republicans are suddenly waking up to the urgent need for tough AI safety rules, a sharp reversal from last year when House Republicans and Chairman Cruz tried to impose a 10-year moratorium on state AI regulations that the tech bros wanted. I welcome the urgency,” Cantwell posted. “Meaningful legislation would require the most powerful AI models undergo testing by scientists and experts at our national laboratories to assess whether they could enable sophisticated cyberattacks or aid the development of biological or nuclear weapons.”

Klobuchar and Thune previously collaborated on the Artificial Intelligence Research, Innovation, and Accountability Act of 2023 that called for strong transparency and risk management reporting requirements from companies deploying AI. It did not pass during that session of Congress but has informed the new bill with Cruz’s involvement, though it is likely to be “a fundamentally different bill,” the Democratic committee aide said.

A markup scheduled before the August recess that would’ve overseen a version of the Thune-Klobuchar bill was cancelled, with Cruz saying that he wanted to focus on child safety legislation.

Cantwell’s stance comes as the public discourse on AI has centered on allegations from a former employee of both OpenAI and Anthropic who said that neither company is acting responsibly while developing advanced AI. In response, scores of lawmakers issued calls for the federal government to effectively regulate the emerging technology.

]]>

CIA feared terrorists could exploit Y2K glitches, declassified briefs show

David DiMolfetta — Fri, 11 Sep 2026 13:37:00 -0400

Newly declassified CIA records detail how analysts viewed emerging technology threats before 9/11, from fears that terrorists could exploit Y2K computer glitches to assessments that extremist websites were helping recruit fighters.

The CIA released the records — President’s Daily Brief assessments prepared for the president and senior advisers — to mark the 25th anniversary of the Sept. 11, 2001 attacks. The collection also includes assessments describing how militants used the internet to circulate chemical and biological weapons information.

In a Dec. 30, 1999, assessment issued during President Bill Clinton’s administration, analysts warned that computer failures accompanying the new year could create opportunities for terrorists.

The fears around a mass Y2K problem stemmed from computer systems that recorded years using two digits, raising concerns that they — and the many crucial parts of American life they underpinned — would misinterpret the transition from 1999 to 2000 and malfunction. The CIA’s assessment anticipated that such disruptions could create opportunities for attackers. The agency did not identify a specific hacking operation or establish that one occurred.

“Computer degradations due to Y2K glitches are likely to facilitate opportunistic attacks by groups or individuals, including hackers who may be affiliated with terrorist organizations,” the assessment said.

“Bin Ladin affiliates arrested this month in Pakistan, for instance, are linked to the extremist Muslim Hackers Club,” it added, using an alternate spelling of the al Qaeda leader’s name. Multiple U.S. agencies had issued warnings about the hacking group in the months following the 9/11 attacks.

The assessment’s findings came amid heightened concern about attacks surrounding millennium celebrations. Authorities that month arrested Ahmed Ressam while he was bringing explosives into the United States from Canada, disrupting a plot to bomb Los Angeles International Airport, according to the 9/11 Commission Report.

The commission’s 2004 findings also noted that officials had feared terrorists would take advantage of Y2K breakdowns. The newly released record provides a specific example of that concern, including the alleged hacker connections. Its unredacted text does not explain the nature of those links or the group’s cyber capabilities.

By August 2000, analysts were describing how extremist websites were helping draw recruits into a network that bin Laden could tap for terrorist operations.

The releases, while they appear to not contain revolutionary new findings, highlight how U.S. spies were assessing how the then-emerging digital age would augment terrorist groups’ ability to recruit followers, share information and possibly plan digital attacks.

The documents also offer a glimpse of how intelligence judgments were presented to the White House at the time. The released passages generally do not assign explicit confidence levels to their conclusions, unlike many intelligence assessments produced today that use them to convey how strongly the available evidence supports analysts’ judgments. Redactions further limit what the public can determine about the sources behind those findings.

“The products trace the story of CIA analysts’ evolving understanding of al-Qa'ida and efforts to highlight and warn of Usama Bin Ladin’s attack plotting despite sparse, vague, and imperfect information,” the agency said in a statement.

The assessments from that time also examine how conflict elsewhere in the world was driving resources to al Qaeda.

Chechnya, a predominantly Muslim region in southern Russia, fought for independence from Moscow after the Soviet Union collapsed. Russian forces fought separatists there in 1994 to 1996 and launched another war in 1999. The conflict also attracted Islamist militants from abroad, and CIA officers were tracking how recruitment for that fighting could benefit bin Laden’s network.

An Aug. 17, 2000, assessment, also from Clinton’s presidency and titled “Recruits for Chechnya Swelling Bin Ladin’s Ranks,” said worldwide recruitment for the fighting in Chechnya had produced a surplus of militants that bin Laden and other extremists were drawing upon to promote terrorism abroad.

Supporters were recruiting young Muslims in Britain, Pakistan, Turkey, the Arabian Peninsula and elsewhere for training at camps affiliated with bin Laden in Afghanistan, the assessment said.

“Fueling the recruiting are Islamic extremist Internet sites that glorify the mujahidin in Chechnya, highlight Russian brutality, and stress that fighting in a jihad is every Muslim’s religious duty,” analysts wrote. The passage does not identify the websites or quantify how many recruits were attracted.

Another assessment, dated Feb. 13, 2001 — less than a month after President George W. Bush took office — described militants gaining access to crude chemical and biological production information and training through an international network based in Afghanistan that partly involved the use of the web.

“The network uses training camps, electronic media, and the Internet to disseminate CB data,” the document says, using an abbreviation for chemical and biological information.

The records offer a fuller view of the intelligence reaching the White House in the years before 9/11, showing how analysts described bin Laden’s ambitions and possible attacks.

The release also comes as the intelligence community marks a quarter-century of changes in how it gathers information, shares warnings and tracks threats. In interviews with Nextgov/FCW, former senior intelligence officials reflected on the changes that followed 9/11 and the lessons they believe still matter today.

]]>

25 years after 9/11, spy agencies face old lessons and new threats

David DiMolfetta — Fri, 11 Sep 2026 10:22:00 -0400

As Air Force One returned to Washington, D.C. on the night of Sept. 11, 2001, Michael Morell looked out a window and saw an F-16 off the aircraft’s wingtip. Beyond it, the Pentagon was still smoldering.

Morell, then-President George W. Bush’s intelligence briefer, had spent the day answering the president’s questions and relaying intelligence as the country tried to understand the attacks that had hit the Pentagon and the World Trade Center in New York City. A quarter-century later, that view from the window remains one of his most vivid memories.

At the National Security Agency, Rick Ledgett recalled colleagues weeping in the hallways as the agency shifted into crisis mode. Within a month, it had transferred 3,000 analysts to counterterrorism work, he said.

“Personally, it was shattering,” Ledgett, who later served as NSA deputy director, told Nextgov/FCW in an email, adding that he “saw people step up and perform in ways they hadn’t before.”

The response to the worst terrorist attack in modern history would forever transform the government they served. The efforts involved reorganizing intelligence agencies, expanding surveillance powers and drawing spies deeper into a global campaign to capture or kill suspected terrorists. It also produced longstanding debates over torture policies, targeted killings, mass surveillance, prolonged wars and disputes over presidential power that became inseparable from efforts made to keep Americans safe from another catastrophe.

Twenty-five years later, former officials describe a country better prepared to disrupt terrorist organizations, but facing a broader mix of threats from foreign governments, hackers and extremists able to reach Americans online.

“I think we’re much safer than we were on September 10, 2001, from the threat of terrorism,” Morell, who later served as CIA deputy director and acting director, told Nextgov/FCW in an interview. But threats from nation-states like China or Russia, he said, are the most significant the country has faced since the Cold War.

The 9/11 Commission — chartered by Congress and the White House in late 2002 to present a full accounting of the attacks and response — described in its final report failures spanning imagination, policy, capabilities and management. Agencies held pieces of information that did not come together, and domestic defenses were not effectively mobilized despite mounting warnings about al Qaeda.

Just over a month before the attacks, Morell briefed Bush on a landmark intelligence report titled “Bin Ladin Determined To Strike in US” that flagged “patterns of suspicious activity in this country” consistent with preparations for hijackings and other attacks. He told Nextgov/FCW it outlined al Qaeda’s intent to attack the country but did not specifically identify when, where or how a specific plot would unfold.

Getting agencies to work together

The transformation that followed 9/11 sought to make cooperation a standing responsibility. Congress created the Office of the Director of National Intelligence in late 2004, establishing a lead unit responsible for integrating intelligence across agencies that collect data from human informants, satellite imagery, intercepted phone calls and other clandestine sources. The Department of Homeland Security was also stood up in late 2002 to drive domestic defense efforts.

The expansion also deepened agencies’ reliance on contractors. After years of workforce drawdowns following the Cold War, the federal government turned to private companies to meet demands for intelligence collection, analysis and technology support, according to a 2015 report from the Congressional Research Service. Today, they play a ubiquitous, outsized role in America’s national security missions.

Lauren Goldman, a former official in ODNI’s Cyber Threat Intelligence Integration Center and National Intelligence Council, recalled a major shift in the basic expectations that governed intelligence-sharing across government and the private sector.

“The burden became on why you shouldn’t share versus why you should share,” she said. That culture has permeated many categories of today’s national security landscape, especially cybersecurity.

ODNI sought to help ensure information reached people who needed it, including officials who might otherwise have been left outside an agency’s reporting channels, Goldman said. Because the office does not collect intelligence itself, it can assess reporting without the same institutional or cultural attachment to a particular collection method, she added.

James Clapper, who served as director of national intelligence for more than six years under the Obama administration, said that coordination requires a “full-time champion.” Under pre-9/11 arrangements, the CIA director also headed the intelligence community, but agency demands often consumed the attention needed for that broader role, he said.

Clapper said the DNI position was created to coordinate intelligence agencies but was not given full authority over them. Entities like the State Department or the Pentagon, for instance, retained control of their own intelligence shops. Still, the DNI could influence their priorities through oversight of intelligence funding and a direct advisory role to the president, he said.

“I do think there’s a need for an overall champion of collaboration and integration,” Clapper told Nextgov/FCW. His remarks come as the Trump administration has sought to shrink ODNI’s workforce and consolidate its functions, on grounds that the office has become bloated and duplicates work performed elsewhere in the intelligence community.

Ledgett credited ODNI with promoting cooperation, as well as running the National Counterterrorism Center and other hubs. But he said it had struggled to take power from individual agencies, particularly the CIA, and secure the president’s attention.

Morell said the CIA knew al Qaeda wanted to attack the United States but failed to uncover the 9/11 plot. He blamed that failure largely on a shortage of resources to gather intelligence, rather than agencies failing to share what they knew. That view differs in emphasis from the commission’s account, which identified missed opportunities to share or act on information.

Clapper said mistaken assessments about Iraq’s weapons of mass destruction programs also reshaped how spy agencies worked. The Bush administration cited those assessments in making its case for its 2003 invasion. Clapper, who had involvement in a key flawed assessment of Iraq’s WMDs, said the failure pushed agencies to scrutinize their sources and challenge assumptions more rigorously. U.S. analysts, in particular, had relied heavily on an Iraqi informant whose claims came through German intelligence, without direct access to question him.

The limits of intelligence powers

Global surveillance became one of the defining disputes of the post-9/11 era. The NSA’s bulk collection of domestic telephone records, exposed in Edward Snowden’s 2013 disclosures, drew scrutiny over both its intrusions into Americans’ lives and its value as a counterterrorism tool.

In 2014, the Privacy and Civil Liberties Oversight Board recommended ending that collection program, and a federal appeals court ruled in 2015 that it exceeded what Congress had authorized under Section 215 of the Patriot Act. But other surveillance programs exposed by Snowden — including those operating under the contentious Section 702 authority — have remained central to U.S. intelligence-gathering.

Ledgett, who led the NSA’s assessment of the damage from Snowden’s disclosures, defended the post-9/11 surveillance effort as “legal and authorized,” though he noted the intelligence community could have done better at public relations.

The incident forever changed how lawmakers and journalists examine spy agencies’ collection activities and whether Americans’ privacy is adequately protected. Contemporary mass surveillance debates have taken on new forms, including spy agencies’ purchases of sensitive data from commercial brokers and police use of artificial intelligence-powered tools to search vast networks of information about people’s movements and activities.

The CIA’s detention, interrogation and torture practices produced another reckoning. The Senate Intelligence Committee’s 2014 investigation concluded that the agency’s coercive interrogation techniques were ineffective in obtaining intelligence or cooperation and that the CIA impeded oversight and misrepresented aspects of the program.

Morell, who has previously disputed the Senate report’s conclusions, argued that responsibility also belonged to the president and other authorizing officials, and he rejected the idea that the agencies had acted entirely on their own.

Asked whether the country went too far in its broader response in the War on Terror, Morell said that “as a nation, we overreacted,” pointing mainly to the Iraq War and the length of the war in Afghanistan.

Intelligence agencies also helped identify and track suspected terrorists for controversial targeted killings, including those carried out through U.S. drone strikes. The debate reached American citizenship itself when the government targeted Anwar al-Awlaki, an American and al Qaeda operative, in a lethal strike. Defending the action in 2013, then-President Barack Obama argued that citizenship could not shield an operative plotting attacks when capture was not feasible.

Today, drones have become a battlefield mainstay in Russia’s war against Ukraine, and Western militaries are studying how to build, deploy and counter them. Weapons that drew scrutiny for their role in U.S. counterterrorism operations are now also held up as examples of military innovation, with NATO learning directly from Ukrainian drone operators. But questions about whom they kill and how they are used persist, as the United Nations documents their growing toll on civilians.

The fight over intelligence powers also became increasingly partisan. President Donald Trump’s disputes with intelligence officials over Russia’s interference in the 2016 election helped fuel his accusations that agencies were being used against him. A bipartisan Senate investigation upheld the intelligence community’s finding that Moscow sought to help him win. But allegations of political “weaponization” became a recurring theme in Trump’s criticism of intelligence and law enforcement agencies.

A changing threat landscape

Intelligence agencies now face a wider range of threats, including foreign hackers with access to fast-evolving AI tools.

Morell said the United States learned how to degrade terrorist organizations by denying them safe havens where they could train, raise money and prepare attacks. But the growing reach of digital networks has created other ways to threaten the U.S. without physically entering it.

Goldman pointed to critical infrastructure, where connected systems can allow the effects of an intrusion to spread beyond a single building or city, as a target of such threats. Hardening airports and other physical targets has value, she said, but does not eliminate vulnerabilities created by that cyber interconnectedness.

State-backed hacking illustrates the stakes. About six years ago, Russian cyber operatives compromised SolarWinds software as part of an espionage campaign that infiltrated multiple federal agencies and some 100 private companies. China-linked Salt Typhoon hackers later penetrated major telecommunications networks, targeting senior officials’ communications and accessing systems used for court-authorized wiretaps. And another Chinese campaign, Volt Typhoon, embedded hackers in American critical infrastructure, with U.S. officials warning that the access could enable disruption during a future conflict.

Terrorist groups have also found new ways to spread propaganda and reach potential recruits online. The Digital Citizens Alliance and cybersecurity firm risk3sixty recently examined 25 piracy-based internet television services and identified terrorist-linked broadcasters on 17 of them. Hezbollah’s Al-Manar satellite television station appeared on all 17, according to their analysis.

Some terrorism-prevention measures have also lapsed. A Sept. 8 Government Accountability Office report found that the 2023 expiration of the Chemical Facility Anti-Terrorism Standards program ended continuous screening of roughly 500,000 people for possible terrorist ties. The program covered about 3,200 high-risk chemical facilities, whose owners cannot independently access federal terrorist watchlist information.

As officials who lived through 9/11 reflect on their time in government, they want the next generation to remember how warnings were missed and what it took to get agencies working together.

Goldman said analysts need to keep questioning their conclusions, examining blind spots and sharing information, despite the demands of daily work. Clapper emphasized the difficulty of persuading policymakers to act on a danger the public has not yet experienced.

Ledgett expressed confidence in those who will take their place.

“Everyone eventually retires, and organizations continue,” he said. “The next generation steps up and delivers when needed.”

The U.S. is skilled at responding once a crisis arrives, but has struggled to prepare before it does, Morell said.

“Becoming a nation, a White House, a Congress, an American people that is proactive in preventing things — preventing bad things — is much better than only being reactive to them,” he said.

]]>

On 9/11’s 25th anniversary, modernizing government starts with remembering our past

Kshemendra Paul — Fri, 11 Sep 2026 09:00:00 -0400

As we mark the 25th anniversary of September 11, 2001, we honor those we lost and reflect on progress in our national security and public safety architectures. Yet behind the ceremonies lies a persistent, quiet vulnerability. Today, public sector Chief Data Officers and Chief Data and AI Officers have inherited the imperative to connect the dots, share information across boundaries, and better serve Americans.

Too often, however, they stumble over the same structural barriers that blinded us a quarter-century ago: isolated data silos, stove-piped legacy systems, failures of intergovernmental coordination, and interagency turf wars. Worse, threats have accelerated, driven by transnational crime, rapid technological advances in artificial intelligence and cybersecurity, and geopolitical conflict. These barriers are compounding into AI data readiness gaps. The cost of forgetting is not theoretical.

Following the 9/11 Commission’s diagnosis, legislation informed by the Markle Foundation required the President to establish a cross-domain, whole-of-government Information Sharing Environment (ISE). The ISE attributes were forward-leaning. They anticipated current efforts to catalog data, build trust architectures, and automate discovery. Sharing and safeguarding were treated as aligned, with guardrails for sharing allowing for more lasting, efficient, and secure impact. The underlying approaches were a data-first approach to architecting enterprise transformation and not specific to terrorism-related information.

Through ISE-linked governance, leaders established policies and technologies that strengthened secure information sharing. These efforts contributed to nearly $531 million in savings and allowed the Government Accountability Office (GAO) to remove nationwide terrorism-related information sharing from its High-Risk List in 2017. In particular, GAO highlighted leadership commitment among federal, state, local, tribal, international, and private sector partners as a key driver for the collective results.

Yet, a subsequent GAO review underscored that progress stalled as dedicated coordination, partner engagement, investment, and oversight were allowed to lapse. At the 2022 National Summit on Intelligence, hosted by the Department of Homeland Security and the International Association of Chiefs of Police (IACP), more than 100 frontline leaders warned that without active governance, cooperation has drifted. Why are we still fighting these battles? We must use this anniversary to pause, remember, and build upon our past successes.

Bridging Federal, State, and Local Agencies

Like many others, I decided to be part of the solution post-9/11. I first served as the Department of Justice’s Chief Enterprise Architect and delivered the National Information Exchange Model (NIEM) into use across all 50 states. I then served as the Federal Chief Architect at the Office of Management and Budget. Subsequently, I served for six years as the Presidentially-designated Program Manager for the Information Sharing Environment (PM-ISE) and governmentwide lead for information sharing.

In that role, I saw firsthand the friction of cross-agency information sharing. Bridging the gap between domestic criminal intelligence and foreign intelligence meant overcoming cultural aversion and perceived risk. There was always tension and pushback from powerful agencies that preferred to operate unilaterally. I sat in rooms where the right answer was obvious and the institutional will to act on it was not.

Acting as an empowered convener backed by the White House, the PM-ISE championed the collaborative, collective approaches that were otherwise impossible. Co-chairing the National Security Council’s (NSC) information sharing and safeguarding inter-agency policy committee, we developed and oversaw implementation of our national strategy. We embedded robust operational policies to protect privacy, and civil liberties—protections that were operationalized nationally and later affirmed in federal courts.

The Operational Impact of Federated Governance

Federated, transparent, participatory, and collaborative governance was the true engine of this progress. It worked because it aligned incentives rather than imposing control: rather than forcing local agencies into a monolithic federal database, we treated state, local, and private sector leaders as first-rank partners through interlocking advisory bodies.

We built upon and integrated successful models like the FBI’s Criminal Justice Information Services (CJIS) division and its Advisory Policy Board. To bridge information sharing architectures, we partnered closely with industry and standards organizations. That collaboration continues today with NIEMOpen.

A core success of the ISE was delivering the Nationwide Suspicious Activity Reporting Initiative into operational use. This standardized tips and leads, used the national network of fusion centers as local filtering, analysis, and information sharing nodes, and successfully integrated a million frontline officers into a national architecture protecting Americans.

To make federated governance tangible for those frontline officers, we championed operational coordination and deconfliction between High Intensity Drug Trafficking Areas (HIDTA), Regional Information Sharing Systems (RISS), and the National Network of Fusion Centers. While dedicated professionals continue to share tactical intelligence locally, state-wide and regionally, and with federal partners daily through mechanisms such as Joint Terrorism Task Forces, it is the enterprise-level strategic governance required to scale these efforts and connect the dots nationally that has atrophied.

Fortunately, the DOJ’s Global Justice Information Sharing Initiative and its Criminal Intelligence Coordinating Council (CICC) have just been formally rechartered. We have a critical window to restore them to their capstone role with our public safety and domestic law enforcement architecture.

The Thread Forward: Evidence, AI, and Trust

The ISE structural reforms directly connect the hard-earned lessons of post-9/11 counterterrorism to the modern mandates of the Foundations for Evidence-Based Policymaking Act, and the urgent need to counter AI-accelerated threats. Today, semantic interoperability is more critical than ever. As NIEM has evolved into NIEMOpen, it restricts governance overhead to the minimum common standards needed across a mission ecosystem, providing the crucial semantic foundation data leaders need.

If we attempt to fuel modern AI tools using the siloed, non-interoperable data architectures of the past, we risk billions in failed IT modernization and dangerous algorithmic hallucinations. Federated and open frameworks like NIEMOpen are not just administrative best practices; they are the semantic prerequisites for trustworthy and interoperable AI.

The Call to Action

On this 25th anniversary, our nation faces a clear choice. To secure our communities and lead in the AI age, the Administration and Congress must take decisive action:

Fully Restore Global and the CICC: Fully empower the newly rechartered Global advisory committee and its CICC. For a local police chief, this federated governance means receiving a single, actionable threat picture rather than drowning in duplicative, siloed federal alerts with resulting likely gaps.

Reconsider the PM-ISE: Congress and the Administration must resolve statutory ambiguities to formally re-establish and empower the Program Manager. Anchoring this role as in the past with the NSC and OMB will provide the independent leverage needed to align federal equities, integrated state and local needs and perspective, and refresh and broaden our national strategy.

Embrace Federated Data Governance for AI: Follow the principles of the Evidence Act and NIEMOpen. We must treat data as an enterprise asset governed by shared semantic standards and robust safeguards. To lead this effort, OMB should lean on agency CDOs to refresh the Federal Data Strategy and consider use of NIEMOpen as the basis for inter-agency and state and local data sharing standards. ensuring our AI models generate trustworthy insights.

The technology of government matters, but it is people, trust, and federated governance that make it work. By bringing back the human and structural DNA of the ISE, we can honor the legacy of 9/11 with enduring institutional renewal.

Kshemendra Paul retired from federal service in 2025. He is recognized for his pioneering results and solutions to improve government services and operations. He started his career in the private sector as technology leader and entrepreneur. In the second half of his career, he served in a variety of federal agencies and the White House in leadership roles such as assistant inspector general, governmentwide lead for information sharing, federal chief architect, and chief data officer. His focus now is public policy advocacy. He also advises leaders and organizations on data and artificial intelligence strategy and governance.

]]>