Nextgov/FCW - Artificial Intelligence

US opposes global AI standards but sees value coordinating on ‘real-world harms,’ State official says

Edward Graham — Tue, 02 Jun 2026 16:37:00 -0400

Although the Trump administration is opposed to working with international governing bodies to establish any frameworks around the use and development of AI technologies, a top State Department official said conversations are still taking place with allies about coordinating responses to certain national security threats.

During an Atlantic Council event on Monday, Under Secretary of State for Economic Affairs Jacob Helberg said the White House “is highly skeptical of supranational bodies in the business of governance,” but added that the administration is open to closer global collaboration on other tech- and cyber-related issues despite its “America First” agenda.

“There is a conversation surrounding the benefits — the potential benefits — of coordinating with other countries on addressing, you know, cybersecurity threats, threats around physical infrastructure risks, around deepfakes,” Helberg said. “So, you know, coordinating with other partners on the identification of real-world harms that are worthy of coordination.”

He added that it “is a conversation that's very much taking place, but as an administration, we haven't yet adjudicated on, you know, what the final road ahead lies on that front.”

President Donald Trump has strongly pushed back on U.S. engagement with international bodies like the UN, including directing officials to withdraw from several global bodies and slash funding for other entities.

This opposition to international collaboration has extended to the nation’s AI policies, with the White House favoring continued support for American technology dominance over giving transnational organizations any authority to craft global standards that can be shaped by U.S. adversaries like Russia and China.

The third pillar of the administration’s AI Action Plan, released in July 2025, focused on “lead[ing] in international AI diplomacy and security,” with the document voicing opposition to international bodies creating any AI governance frameworks.

“The United States supports likeminded nations working together to encourage the development of AI in line with our shared values,” the plan said. “But too many of these efforts have advocated for burdensome regulations, vague ‘codes of conduct’ that promote cultural agendas that do not align with American values, or have been influenced by Chinese companies attempting to shape standards for facial recognition and surveillance.”

Trump also signed an AI executive order on Tuesday that calls for firms to voluntarily provide the federal government with pre-public releases of their models to review them for potential cybersecurity or national security risks.

U.S. officials have also warned European Union allies to refrain from further regulating AI technologies and other emerging capabilities. During an April event in Brussels, Helberg said the bloc’s current regulatory regime is pushing American companies out of its orbit and that "Europe is accruing a [technology] lag that will not be reversible in years.”

]]>

Anthropic held cyberthreat briefings with agency CIOs last month

Alexandra Kelley — Tue, 02 Jun 2026 15:29:00 -0400

Leading artificial intelligence developer Anthropic hosted briefing sessions for federal agency chief information officers in early May, several sources familiar with the sessions told Nextgov/FCW.

Meetings occurred May 7 and May 8. While briefing topics varied, they focused on defending digital assets from cyber threats powered by advanced AI models including Anthropic’s Mythos Preview, the sources said.

Other branches of government have been informed of Mythos’s capabilities. In mid-May, lawmakers on the House Homeland Security Committee received a briefing with Anthropic executives on Mythos’s ability to detect software vulnerabilities.

The launch of Mythos Preview in early April came alongside Anthropic’s announcement of Project Glasswing, an initiative that granted access to the model in its beta form to multiple participating private sector partners. On Tuesday, Anthropic announced its expansion of Project Glasswing to include roughly 150 new partners following initial feedback from inaugural companies.

New participants in Project Glasswing are from industries that weren’t included in the first cohort, with sectors like power, water, healthcare, communications and hardware now part of Anthropic’s initiative.

Project Glasswing’s debut came just weeks after the Pentagon labeled Anthropic a supply chain risk in response to the company contesting the use of its technology in Pentagon operations with autonomous weaponry and American surveillance. The designation prompted President Donald Trump to order the government to halt all use of Anthropic products. The legality of the supply chain risk designation is being contested in court following Anthropic’s lawsuit against federal agencies and their leadership.

Despite the supply chain risk designation, the federal government is keen to understand Mythos’s threat capabilities. A long-awaited executive order on AI was slated to address how the federal government analyzes AI-driven cyberthreats, including granting intelligence and security agencies access to advanced frontier AI models, but signing of that order was postponed after Trump expressed doubts that it might hinder AI innovation. Trump signed a scaled-down version of that order Tuesday that implemented a lesser degree of federal oversight on such advanced models.

]]>

Trump signs AI executive order after postponement last month

Alexandra Kelley and David DiMolfetta — Tue, 02 Jun 2026 13:02:00 -0400

President Donald Trump on Tuesday signed a cybersecurity-focused artificial intelligence executive order directing national security and civilian agencies to expand oversight of advanced AI systems, marking the administration’s latest attempt to balance growing fears over catastrophic AI-enabled cyber risks with a broadly pro-innovation agenda.

The directive scales down the degree of federal oversight of AI models from what was initially included in an earlier version that was set to be signed two weeks ago, but that signing was postponed amid overregulation concerns from industry.

Per the order, companies developing cutting-edge AI systems would be encouraged to provide the federal government with 30 days of pre-public access to those models, as well as limited early access for select critical infrastructure operators. An earlier outline of the order viewed by Nextgov/FCW suggested the government would be granted a longer window of 90 days to assess covered frontier models.

The more limited pre-release access period, coupled with language in the order that explicitly prohibits licensing or preclearance requirements, suggests the administration is seeking visibility into advanced AI systems without establishing a formal approval process before companies can release new models, a dynamic that is more favorable to industry.

One section of the order, focused on cybersecurity, directs federal agencies to secure Defense Department and other national security networks within 30 days. Another includes a binding operational directive to secure federal civilian networks and facilitate access to frontier AI models across critical infrastructure sectors, including hospitals, banks, utilities and state and local governments, which must also be issued within 30 days.

It also calls for the Treasury Department — with support from the Office of the National Cyber Director, the National Security Agency, the Department of Homeland Security and the Cybersecurity and Infrastructure Security Agency — to establish a voluntary coordination clearinghouse between the government, AI companies and critical infrastructure operators.

Additional provisions would direct the Office of Management and Budget to identify existing federal grant funding that could support AI vulnerability-detection efforts within 30 days. It also tasks the Office of Personnel Management with increasing cyber hiring via the U.S. Tech Force within 60 days. The Tech Force, launched in December, has expressly been recruiting cyber talent for the last several weeks, though it has only onboarded 10 total employees thus far.

The other section of the directive focuses on establishing a new government framework for overseeing advanced AI systems, including the creation of a classified benchmarking process to determine which models qualify as “covered frontier models.”

Per the order, the White House Office of Science and Technology Policy, the National Institute of Standards and Technology, CISA and others would have 60 days to establish the classified evaluation process.

The NSA, in consultation with those agencies, would then be tasked with formally determining which AI systems meet the threshold. The NSA’s involvement in these efforts was reported in May by Nextgov/FCW.

In the same section governing frontier model development, the Commerce secretary is also tasked with assisting in the development of a classified AI benchmarking process that will inform the voluntary framework for AI developers. The final draft of the order states that the agency's secretary will work “through the Director of the National Institute of Standards and Technology,” a caveat that wasn’t included in the initial draft, per a copy reported last month by Politico.

The administration’s approach to AI has shifted in recent months amid the emergence of Anthropic’s Mythos, a powerful cybersecurity-focused AI model that has become a major driver of government discussions, as officials grapple with how advanced AI systems can rapidly uncover vulnerabilities across computer networks.

OpenAI’s recent release of GPT-5.5-Cyber, which also demonstrated sophisticated cyber capabilities, has further heightened concerns in Washington over how quickly these systems are advancing and how they could reshape both cyber defensive and offensive operations.

]]>

Argonne launches high-performance computing-backed AI research service

Alexandra Kelley — Wed, 27 May 2026 17:02:00 -0400

Argonne National Laboratory announced on Tuesday that it launched a new platform to offer researchers access to various artificial intelligence models, the latest move supporting the Department of Energy’s mission to spur advanced research and innovation in AI.

The lab is deploying an AI inference service — a cloud-like offering that is designed to analyze data, make connections and supply predictions — to facilitate scientific access to leading AI models. The service will provide an array of large language models and scientific foundation models to users in the national lab apparatus.

“Our inference service helps close the gap between developing AI models and putting them to work in scientific research,” Michael Papka, the director of the Argonne Leadership Computing Facility, said in a press release. “By offering AI inference as a shared resource, we enable researchers to apply AI at scale to their data, simulations, and experiments, without the burden of building and maintaining their own infrastructure.”

Hardware powering the inference service is headquartered within Argonne. Leveraging the lab’s flagship exascale computer, Aurora, the inference service will also run on Argonne’s NVIDIA DGX A100 cluster, Sophia, along with the ALCF’s SambaNova SN40L chip cluster, Metis.

The models offered via Argonne’s inference service — which include commercial and in-house options — are pre-trained. Granting researchers facilitated access to powerful, tailored models will help them “spend less time managing models and more time testing hypotheses,” said Venkat Vishwanath, AI and machine learning lead at the ALCF.

The current models available include open-weight models, domain-specific science foundation models, among others, Papka told Nextgov/FCW.

“Instead of taking days or weeks to analyze data, scientists can rapidly interpret results, refine experiments and explore complex systems in ways that weren’t practical before,” Vishwanath said.

The new offering is based on a 2025 paper that outlined a framework “to give researchers the ability to run multiple AI tasks in parallel on different models without relying on commercial cloud services.”

This effort contributes to Energy’s ongoing Genesis Mission, a project that aims to spur advanced research and innovation in AI by leveraging federal datasets and resources.

Per the press release, researchers at Los Alamos National Laboratory, Brookhaven National Laboratory, Lawrence Berkeley National Laboratory, Fermi National Accelerator Laboratory, Lawrence Livermore National Laboratory, Oak Ridge National Laboratory, Sandia National Laboratories and Thomas Jefferson National Accelerator Facility are also able to access the inference service.

The service can leverage its model access to work beyond research in AI. Although it functions as a new pillar in the Genesis Mission, ALCF says that the inference service can be applied to other fields, such as research in fusion energy, chemistry and materials science.

]]>

How the Library of Congress is using both AI and volunteers to unlock public broadcasting history

John Breeden II — Fri, 22 May 2026 17:42:00 -0400

Public broadcasting has a long history of capturing important moments in American life. It preserved voices from the civil rights movement, debates over war and foreign policy, regional arts coverage and local public affairs programs that reflected the people and places shaping the nation. But many of those moments have also been hard to find, buried in tape vaults, archives and library collections that few people would ever be able to search or even really know about.

That is part of what makes the American Archive of Public Broadcasting (AAPB) so interesting. A collaboration between GBH, which is the Boston public media organization formerly known as WGBH, and the Library of Congress, the archive is working to make historic public media more searchable and accessible, in part by using AI-generated transcripts as a starting point.

The public-facing correction layer for that effort is called FixIt+, a volunteer platform where people can review and refine machine-generated transcripts from older radio and television programs. As AAPB Archives Outreach Manager Meghan Sorensen explained in an interview published by the Library of Congress, “FixIt+ is a volunteer transcript correction platform and open-source project maintained by our team at GBH for the American Archive of Public Broadcasting. Its mission is to make historic public media more accessible by inviting the public to help update and correct computer-generated transcripts in a way that feels easy and engaging.”

That approach makes a lot of sense. I work with transcripts fairly often myself. When I am covering an important speech, a major announcement or a policy presentation, I will often check the transcript as I listen so I can catch words or details that went by too quickly. For modern events, AI-generated transcripts are usually pretty good. Even so, they still stumble in predictable ways. Laughter, coughing and side comments can confuse them, and they sometimes force nonverbal sounds into words that were never spoken.

That problem becomes much more obvious when the recordings are older. In trying out and working with FixIt+, I spent time with broadcasts from the 1960s and 1970s, and the limitations were easy to hear. The audio may have been broadcast quality for its time, but by modern standards it can sound thin, noisy or compressed. Regional accents can make the software hesitate, and background sounds only make the job harder. If the archive simply accepted the AI-generated text as final, there would almost certainly be mistakes left behind. I found quite a few pretty obvious ones within the first several minutes of using the platform. They were not huge errors, but for important historical moments, the transcripts should be as accurate as possible.

FixIt+ handles that problem in a practical way. As I listened to audio or watched the old television program, I could type my suggested correction directly into the line on the transcript. Once I saved the change, it became part of the archive workflow so that other volunteers could review what I had done. They could then approve my change or suggest an alternative. Only after a transcript reaches volunteer consensus is it treated as complete. The project describes this as a “human-in-the-loop” process, meaning people improve transcripts generated by computers instead of relying on the software alone. Sorensen put the larger point plainly: “Technology gives us a great jumping-off point, but it is our volunteers who make the real difference.”

What makes the project especially compelling is the material itself. The recordings available for correction are not filler. They include voices tied to civil rights history, national security, foreign policy and regional cultural life. A volunteer might spend time with a May 28, 1961 program involving Freedom Rider Mary Jean Smith, an April 10, 1975 Bill Moyers Journal conversation with former Defense Secretary Clark Clifford about Vietnam and its aftermath, a February 24, 2012 talk by Donald Rumsfeld at Fort Leavenworth, or a January 23, 2004 episode of Black Horizons that includes a discussion of a Buffalo Soldier stage production. A lot of the programs feature very serious topics with rich historical value. And the sheer range of subjects makes it clear that correcting these transcripts is not just a technical chore. It’s a way to help preserve and open up pieces of our nation’s historical record.

The scope of the archive helps drive that point home. The American Archive of Public Broadcasting draws from more than 100 contributing collections, including radio and television stations and other organizations such as WGBH, WNET, Maryland Public Television, Pacifica Radio Archives and the Library of Congress itself. That breadth means volunteers are not working on one narrow slice of programming. They are helping improve access to a wide cross-section of American public media history. The archive notes that transcripts make programs more searchable and usable, while Sorensen explained why that matters in the clearest possible terms: “Without transcripts, much of our catalog remains hidden. With them, the archive becomes a living, interactive resource which can be discovered, shared and explored by anyone.”

That is what gives FixIt+ its real value. It’s not simply a better way to clean up transcripts. It’s a way to bring more people into the work of preserving and opening up public broadcasting history. For volunteers, the task may begin with correcting a few lines of text. But the larger result is that important voices and moments from the past become easier to find, study and understand, allowing them to escape their vaults and become discoverable once more for future generations.

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

]]>

White House postpones signing of AI executive order

David DiMolfetta — Thu, 21 May 2026 12:14:00 -0400

The White House postponed a highly anticipated signing of an artificial intelligence executive order, according to four people with knowledge of the matter.

The order was expected to task national security and civilian agencies with various steps to shore up federal government network defenses using AI models and, notably, establish a voluntary framework for the government to view AI models ahead of release.

The people spoke on the condition of anonymity to provide the details. It’s not entirely clear why the signing was postponed, though some of the people said that multiple tech CEOs expected to attend the order’s Thursday afternoon signing ceremony could not make it.

President Donald Trump expressed misgivings about the order, particularly on Chinese AI competition, according to an Oval Office press report.

“I didn’t like certain aspects of it, I postponed it. I think it gets in the way of, you know, we’re leading China, we’re leading everybody, and I don’t want to do anything that’s going to get in the way of that lead. We have a very substantial standard on AI, it's causing — it's causing tremendous good, and it's also bringing in a lot of jobs, tremendous numbers of jobs,” Trump said. “Again, we have more people working right now than we’ve ever had. I really thought that could have been a blocker.”

Editor's note: This story was updated to include remarks from President Donald Trump

]]>

Anticipated executive order could give NSA a role in voluntary AI model testing

Alexandra Kelley and David DiMolfetta — Wed, 20 May 2026 10:31:00 -0400

White House officials are planning a provision in a forthcoming artificial intelligence executive order that would establish a voluntary information-sharing framework between the government and AI developers to facilitate safety testing of AI models before deployment, according to multiple people familiar with the matter.

The National Security Agency is expected to play a key role under the order and would potentially handle classified testing of models offered up by AI labs before those models are publicly distributed, said some of the people. All sources spoke on the condition of anonymity to discuss sensitive details concerning the order, which they said could be unveiled later this week.

The people also cautioned that decisionmaking in the White House is highly fluid and that details and timing around the final version of the directive may change.

The deliberations over a voluntary framework underscore how the White House is trying to balance competing views within the administration, with some officials and allies pushing for stronger AI safeguards and others favoring a more hands-off approach to the technology to encourage innovation, a stance that’s consistent with prior policy actions.

The plans also appear to show that the Trump administration prefers the intelligence community to lead on AI model testing. The Washington Post reported this month that spy agencies and the Commerce Department are at odds over who should handle model evaluation tasks.

An NSA spokesperson referred Nextgov/FCW to the White House.

“Any policy announcement will come directly from the president. Discussion about potential executive orders is speculation,” a White House official told Nextgov/FCW.

Axios first reported details about the order.

Voluntary pre-deployment testing could give government officials an opportunity to evaluate advanced AI models for cyber-related risks before they are broadly released, including whether the systems can assist with vulnerability discovery.

The administration’s approach to AI has evolved with the emergence of Anthropic’s Mythos Preview, an advanced cybersecurity-focused AI model that has become a major catalyst for the discussions, as officials grapple with how powerful AI tools can identify vulnerabilities across computer networks, including critical infrastructure systems. OpenAI also recently released GPT-5.5, a similar model that can swiftly identify vulnerabilities and assist with complex cyber tasks.

AI-focused information-sharing is not an entirely new concept for the administration. President Donald Trump’s AI Action Plan released last summer called for multiple agencies to establish an AI Information-Sharing Analysis Center to promote sharing of AI-related security threat information across critical infrastructure sectors.

Last week, Rep. Jim Himes, D-Conn., the top Democrat on the House Intelligence Committee, said it would be “insane” for U.S. intelligence agencies to not have early access to advanced artificial intelligence models that could be used for hacking and cyberdefense. He added that the Commerce Department should also play a role in the effort.

Editor's note: This article has been updated to include comment from the White House.

]]>

Virtualitics targets public sector customers with OpenAI partnership

Frank Konkel — Wed, 20 May 2026 09:00:00 -0400

Pasadena-based artificial intelligence solutions firm Virtualitics and frontier AI company OpenAI announced a memorandum of understanding Wednesday to jointly work together to improve mission outcomes for customers with complex, critical workloads — including those within the Defense Department and civilian federal agencies.

The collaboration integrates OpenAI’s frontier AI technology with Virtualitics’ agentic AI platform, Iris, which has several customers in defense and among Fortune 500 firms. Virtualitics Chief Product Officer Aakash Indurkhya said the aim is to pair OpenAI’s frontier AI technology with Virtualitics’ expertise in analytics and AI agent development to create more robust AI agents and enhanced scalability for customers in government and regulated industries.

“This partnership with OpenAI is taking their frontier reasoning models and installing them into the AI agents we’re building,” Indurkhya said. “Users are already pining for this.”

In a statement, Virtualitics CEO Michael Amori said the partnership “lets us pair our readiness expertise with best-of-breed models, while maintaining the trust, transparency and rigor our customers require.”

Indurkhya said OpenAI’s models have the potential to enhance work performed under the company’s current contracts. In one example, the company’s platform is used by the U.S. Marine Corps for predictive maintenance and assessing risk around machine components breaking, tying those data sets to resourcing. Partnering with OpenAI, he said, is “unleashing frontier-level reasoning against those types of tools.”

The move comes as the Pentagon embraces AI and agentic AI agents — tools that perform tasks without human intervention at each turn. OpenAI is one of several frontier AI firms to burst into the public sector market since 2024.

Last July, OpenAI, along with Anthropic, Google and xAI, each received $200 million contracts from the Pentagon to supply AI tools and models. Those companies and others have also discounted their software to government customers through deals through the General Services Administration’s OneGov strategy and served as partners through the administration’s USAi platform. Through this partnership, existing Virtualitics customers will have access to OpenAI’s capabilities as they become available to government customers at increasingly higher security networks.

“Organizations operating in mission-critical environments require AI they can trust,” said Andrew Keene, Head of Government Partnerships at OpenAI. “Our collaboration with Virtualitics allows for richer, context-specific results supporting effective use of AI where readiness and accuracy matter most.”

]]>

USDA is using AI — but doesn’t have required controls to manage risks, watchdog finds

Natalie Alms — Tue, 19 May 2026 16:25:00 -0400

The Agriculture Department is using artificial intelligence to identify risks in the supply chain, estimate yearly corn and soybean yields and make recommendations during the permitting process.

But the department doesn’t have all of the required cybersecurity and governance controls to keep that technology in check, according to an inspector general report released last week, which found that Agriculture doesn’t even have a generative AI policy at all.

The department hasn’t fully implemented cyber and risk controls in its AI systems, as required by federal standards, because it has prioritized using AI over setting up controls for the technology. The Trump administration has sought to aggressively roll out AI across the government, in addition to efforts to dominate with the technology on the world stage.

At USDA, AI systems “could be vulnerable and lack critical security controls, leaving the agency susceptible to data breaches or reputational harm” because of the lack of strong governance around the technology, the new report says.

Agriculture hasn’t followed all the risk management and governance controls set in place by the Office of Management and Budget during the Biden administration and modified by the Trump administration. The department has installed a chief AI officer as required, but it hasn’t updated agency policies — or implemented minimum risk management practices for AI systems deemed especially risky, like those that affect civil rights or critical infrastructure.

Almost none of the AI use cases in the department’s fiscal year 2024 inventory had an authority to operate, a formal approval issued for technology systems meant to make sure that the government thinks through the risks associated with different technologies before using them.

That means that management doesn’t have assurance that the department has cybersecurity controls in place, the report says.

The inventory itself may also be insufficient to account for all potential dangers, as the OIG said the department is at risk of shadow AI — technology used by employees that management isn’t aware of or hasn’t approved — creeping across the department, since it relies only on an annual data call for employees to self-report AI that they use.

The watchdog included several recommendations for the department to implement controls and update policies, all of which Agriculture agreed with.

]]>

Advanced AI models bring government to ‘reflection point,’ CIA official says

Frank Konkel — Mon, 18 May 2026 18:00:00 -0400

Advanced AI models with unique hacking capabilities like Anthropic’s Mythos should bring federal agencies that handle some of the government’s most sensitive information to a “reflection point,” according to one of the CIA’s top tech officials.

“I think it is a reflection point and I think people need to view it in that fashion,” said Dan Richard, Associate Deputy Director of the CIA’s Digital Innovation Directorate. Richard spoke on a panel Friday at the Qualys ROCon Public Sector 2026 conference in Tysons Corner, Virginia.

A previous version of the Mythos software was released to a limited group of tech companies in April with much fanfare, due to its ability to detect countless software bugs and defects.

Security researchers and experts reacted with a mix of excitement and caution, with some warning the software could usher in a new era for hackers and lower the barrier to entry for would-be attackers. Mythos and competing models like OpenAI’s GPT-5.5 have forced executive agencies to grapple with their capabilities and prompted emergency briefings for lawmakers.

Richard said he feels “bullish in terms of the opportunities that are out there,” largely because these AI models can help agencies like the CIA deal with the deluge of data they generate and automate responses to potential threats. He likened the current Mythos-driven moment to Ukraine’s response to Russia’s invasion in 2022.

“[Ukraine] had gone through a decade of the Russians infiltrating their networks and having to deal with that implication, but when the Russians attacked in 2022 the Ukrainians were prepared because they understood they couldn’t do it themselves,” he said. “Shoulder-to-shoulder with them were the private sector vendors to support what they were doing and to help what they’re doing.”

Richard said the U.S. government is in the “same position” now, and public-private partnerships will be key to ensuring the nation gets it right.

“80% of our nation’s critical infrastructure is in private sector hands, so there is no solution that does not include private sector partners,” Richard said. “We talk about partnership all the time, but this is really different. This isn’t transactional. This is us, as a country, figuring out with the academic community, with the private sector community and with our public sector partners working together to be able to defeat and take advantage of what I see as an optimal opportunity for the agency, but for the country.”

Joe Kelly, division director of the Applied Research Laboratory for Intelligence and Security at the University of Maryland, said advanced AI models are going to lower the barrier to entry for would-be hackers.

“The real danger when we look at something like Mythos — whether you believe the hype or not — is it certainly creates what we already see with Claude Code, the ability for script kiddies to cause real damage even without knowing what they’re doing,” Kelly said. “It’s going to lift all those. I do worry about the complexity that we’re entering in this era.”

‘It’s moving so fast, it’s scary’

IonQ Chief Information Officer Katie Arrington, who spent most of 2025 serving as the Pentagon’s chief information officer, said the influx of advanced AI tools — and the speed at which they’re emerging — will test government to the extreme. Existing governance requires IT security vulnerabilities be patched within 30 days, and 15 days for vulnerabilities designated “critical.”

“You don’t have time like that anymore,” Arrington said during a panel at the Qualys event. “We’re talking about a tool that can find every vulnerability in seconds on a platform.”

Arrington said these kinds of advanced AI models weren’t a discussion item even 12 months ago. At that time, the Pentagon was just trying to improve the speed that it could bring general AI tools into its networks.

“It’s moving so fast, it’s scary,” Arrington said. “It scares me and it excites me how fast Mythos came alive.”

Qualys CEO Sumedh Thakar said federal agencies may need to take a more proactive — rather than reactive — approach to risk management to deal with the growing range of threats from advanced AI tools. His company is using its AI-powered cybersecurity tools, including TotalCloud, which recently received authorization to operate in the government’s FedRAMP High environments, to allow customers to automate vulnerability patching, reducing some of the manual processes and “dashboard tourism” cyber professionals otherwise deal with.

Thakar said autonomous remediation allows savvy customers to “battle AI with the speed of AI.”

“Now with attackers leveraging AI, as soon as a patch comes out, they can reverse engineer the patch and they can start to figure out the exploit. Your 30 days has become 30 hours, or three hours,” Thakar said. “What we really focus on is to get over the fear of autonomous remediation. It’s not an option.”

]]>

Nearly 3.4M users across government can use AI through OneGov, GSA official says

Edward Graham — Fri, 15 May 2026 14:51:00 -0400

Millions of federal users can now take advantage of artificial intelligence-specific tools offered through the General Services Administration’s OneGov initiative, an agency official said on Friday.

Speaking on a panel at the ACT-IAC Emerging Technology and Innovation Conference, Birgit Smeltzer — director of GSA’s Office of IT Products, IT Category — said “more than 120 orders have been placed against OneGov’s AI offerings, and that has provided this new technology, or availability, to about 3.4 million across government for this particular technology.”

GSA launched OneGov in April 2025 as a way to offer agencies discounted rates on select private sector technologies and software services by treating the government as one customer. Twenty companies, including Microsoft and Adobe, have reached agreements with GSA so far to offer significant cost savings on some of their products.

These deals have also provided agencies and government personnel with the opportunity to onboard new AI capabilities, which GSA officials previously said is helping speed up government use of and experimentation with the tools.

Smeltzer said multiple agencies have already taken advantage of OneGov’s AI offerings, including the departments of Health and Human Services, Veterans Affairs, Transportation and State, among others. She added that AI offerings accessed through OneGov can enhance workforce familiarity with the tools as the government looks to increase adoption of the capabilities moving forward.

“Now, the agency makes it available to you for maybe a limited time, but you're able to use it in your workday, and can see how it can benefit you and get your work done more efficiently — perhaps without losing your job over [using] it,” Smeltzer said.

GSA officials have touted the cost savings associated with using products purchased through the initiative.

“We want GSA not to just be a shared service across government, but a force multiplier across the government,” GSA Deputy Administrator Mike Lynch said Tuesday at the Coalition for Common Sense in Government Procurement Spring Training Conference in Falls Church.

He added that GSA has identified $1.15 billion in savings through the OneGov program through negotiated discounts of a variety of AI and software tools using the collective buying power of the federal government. The program, Lynch said, will continue to mature in the coming year.

Lynch also said acquiring AI at discounted rates achieved through OneGov is an ideal follow-up for agencies that have experimented with AI and large language models through the USAi.gov shared service platform. Several thousand federal employees have used the USAi platform since GSA launched it last August in response to President Trump’s AI Action Plan.

“We want to see where we can add value, and we’re constantly checking in with industry partners and with agencies to ensure we’re providing world-class service,” Lynch said.

In a statement to Nextgov/FCW, a GSA spokesperson said the AI use and cost savings made possible through OneGov “are real, measurable results from unified buying and direct engagement with [original equipment manufacturers].”

GovExec Editor-in-Chief Frank Konkel contributed to this report.

]]>

NIST aims for summer release of AI cyber guidelines

Alexandra Kelley — Thu, 14 May 2026 16:05:00 -0400

The National Institute of Standards and Technology is slated to debut new guidance on artificial intelligence-specific cybersecurity to help mitigate AI-enabled digital threats while maximizing the benefits of safe AI adoption.

Speaking at a Qualys conference on Thursday, Victoria Pillitteri, the manager of the Security Engineering and Risk Management Group at NIST, said she expects a cybersecurity framework profile for AI to debut “sometime this summer” pending agency approval.

The forthcoming draft AI cybersecurity framework is slated to be accompanied by guidance on control overlays — or sets of tailored cybersecurity baselines to manage risks unique to different AI systems — with the help of NIST’s Center for AI Standards and Innovation.

Pillitteri told Nextgov/FCW that her team and CAISI have started to develop a series of overlay guidance focused on cyber threats targeting agentic, predictive and generative AI systems.

“This [administration’s] priority is speed; being innovative and scaling at speed,” Pillitteri said. “So that means everything does have to move faster. We're trying to evolve the way that we develop, maintain and engage with our stakeholders. For our standards and guidelines, we're trying to ensure that we are addressing these critical areas where cybersecurity intersects AI on multiple fronts.”

She said the draft of overlay guidance for predictive AI is expected to arrive this summer, while the overlay guidance on agentic systems is due in late summer to early fall. NIST plans to finalize the guidance by 2027.

“The intention is to issue all of these guidelines sequentially in draft,” Pillitteri said. “This way we can take lessons learned, improvements, revise everything, but yet still get something out quickly, because we realize adoption is happening now.”

As the federal government mulls how best to understand the digital risks posed by increasingly powerful AI models — such as Anthropic’s Mythos Preview model that was unveiled in its Project Glasswing testing initiative — CAISI will play a key role.

Last week, the Department of Commerce announced it secured new agreements between leading AI developers Google DeepMind, Microsoft and xAI to submit their models for testing in CAISI. Researchers at CAISI will specifically look at the national security ramifications of these companies’ frontier AI models.

As of Thursday, the press release announcing the private sector partnership with CAISI has been removed. The Washington Post reported on Monday that a person familiar with the decision said the page was removed because of sensitivities within the White House.

Pillitteri said that while standards disseminated by NIST can’t keep pace with rapidly evolving technologies like AI, the forthcoming guidance aims to offer a fundamental playbook for vendor and supply chain partners to constantly fortify their data security infrastructure.

“It's not a rip and replace,” she said. “It's, ‘How do we augment what we have? How do we adapt the technologies, the architectures, our products, solutions and services, and how do we augment our workforce … to support the future needs that we're building for ahead?’”

]]>

House Homeland panel gets briefing on Anthropic’s Mythos

David DiMolfetta — Thu, 14 May 2026 13:08:00 -0400

Members of the House Homeland Security Committee were briefed Wednesday on Mythos, the Anthropic artificial intelligence model that has drawn vast attention across the cybersecurity community for its advanced hacking capabilities.

Anthropic executives provided the panel with a live demonstration of Mythos, allowing members to see how advanced AI can identify and reason through software vulnerabilities, according to a committee aide who attended the briefing and requested anonymity to communicate details of the demo.

“What we saw reinforced the urgency of ensuring that federal agencies, including our civilian cyber defenders, can responsibly access and deploy the most advanced U.S. models to find and patch vulnerabilities before foreign adversaries or criminal actors exploit them,” said the aide, who noted the briefing was one of the first live demonstrations delivered to Congress.

President Donald Trump is meeting with Chinese President Xi Jinping this week, where AI competition is expected to come up in the discussion. Last month, the White House accused Beijing of attempting to copy components of U.S. AI systems to build similar models of its own through a process known as distillation.

“As the [People’s Republic of China] aggressively works to close the AI innovation gap with the United States, the committee remains focused on ensuring that America’s AI leadership translates into a durable national security advantage, not a temporary lead that adversaries can copy, steal or rapidly commoditize,” the aide added.

The briefing was “productive and focused on a range of AI security and competitiveness issues,” according to a second person familiar with the demo.

Members discussed how the U.S. can preserve its advantage in AI, including maintaining leadership in compute power and preventing China from obtaining advanced chips, said the person, who added that attendees discussed safeguards for advanced AI models and ensuring future systems are developed and deployed safely and securely.

Lawmakers also asked questions about Anthropic’s engagement with the federal government, including whether an ongoing legal dispute over a Defense Department supply chain risk designation against the company is affecting conversations about the use of AI models across federal agencies, including at CISA, which reportedly doesn’t have full access to the model.

The second person did not add additional details about who in the government has access to Mythos, but said “the implications of advanced AI tools for state and local governments and under-resourced critical infrastructure sectors, including water systems, were also discussed.”

Mythos, unveiled last month, was held back from a full public release on the grounds that it could pose national security risks in the wrong hands. U.S. critical infrastructure stakeholders have been vying for access to the tool so it can be run against their own systems to identify and patch previously undiscovered vulnerabilities that could be exploited by hackers.

The Hill first reported news of the demo.

It’s unclear which government agencies have access to Mythos, although multiple reports and people familiar with the matter previously confirmed to Nextgov/FCW that the NSA is among them.

Senate Intelligence Committee Vice Chairman Mark Warner, D-Va., told Politico that he and Chairman Tom Cotton, R-Ark., were briefed on Mythos by Gen. Joshua Rudd, who leads Cyber Command and the NSA, but did not provide further details.

]]>

‘No time to waste’ in prepping governments for AI cyber threats, top Dem lawmaker says

Chris Teale — Wed, 13 May 2026 14:16:00 -0400

The U.S. Senate’s top Democrat called on the Department of Homeland Security last week to better coordinate its response to artificial intelligence-driven cyber threats with state, local, tribal and territorial governments.

Sen. Chuck Schumer, the New York Democrat who serves as Senate Minority Leader, said in a letter to Homeland Security Secretary Markwayne Mullin that the world is “coming to grips” with the fact that AI systems will soon be better than humans at finding software vulnerabilities. Schumer’s letter came after Anthropic announced last month that its Claude Mythos Preview model is “strikingly capable” at finding cybersecurity vulnerabilities.

Schumer called on Mullin and DHS to work closely with other units of government to prepare them properly for those cyber threats from AI. He noted the threats that states, localities and others face, including to their critical infrastructure, and urged the federal government to do more to protect them. AI could be capable of hacking such systems within a year, he said.

“As AI continues its rapid development — including important cybersecurity advances as well as dangerous new hacking tools — it is imperative that all levels of our government have access to this technology so they can prepare before it’s too late. We must beat cyber criminals in the race to defend our most critical systems from AI-enabled hacking or attacks,” Schumer said in a statement accompanying the letter. “There is no excuse for the Department of Homeland Security’s delay in bolstering state and local government cybersecurity capabilities. We must begin this process now — before there are any major disruptions to hospitals or energy grids — or worse.”

In his letter, Schumer asked Mullin to provide information on how DHS will coordinate with SLTT governments and the private sector to conduct risk assessments of critical infrastructure, and share information about vulnerability discovery and response. He also asked Mullin how DHS will work with other governments to provide remediation solutions, facilitate rapid vulnerability patching, offer access to modern testing and evaluation, and advise governments on identifying top AI talent and training to prepare the next generation of tech workers.

Schumer’s letter noted that those questions come on the heels of the federal government pulling funding for the Multi-State Information Sharing and Analysis Center, which he noted was designated in 2010 as the “primary source” for those functions and more. Since the Cybersecurity and Infrastructure Security Agency announced it had pulled funding for MS-ISAC, the center has moved to a membership model.

For their part, officials in President Donald Trump’s administration have promised more information sharing, especially after the release of the National Cyber Strategy in early March. In a previous public appearance, White House National Cyber Director Sean Cairncross said agencies — including CISA as well as the Department of Justice and Federal Bureau of Investigation — were “looking for ways to streamline information sharing from the [U.S. government] side.”

“Often how we know things is super sensitive,” he continued. “What we know is less so. We want to figure out how to communicate that in a helpful, actionable way, as we work through that on the interagency side, with partners on the state and local side.”

Schumer, however, said the decision to cut funding to the MS-ISAC was a poor one given how AI has shifted the threat landscape.

“Given the monumental changes quickly coming to cybersecurity as the result of frontier AI, and the need for organizations to be able to perceive and contextualize risks earlier than ever before, there could not be a worse time to undercut proven, longstanding MS-ISAC processes, procedures, and resources for sharing cyberthreat intelligence with SLTTs,” the letter said.

Schumer asked for a plan for “coordinating our nation’s response to frontier AI-enabled hacking” by July 1, as well as a nominee to lead CISA.

“AI is changing the cyber battlefield fast — and we cannot let hackers get there first,” Schumer said in a statement. “Hospitals, power grids, water systems, schools, elections, and emergency services cannot be left exposed while criminal gangs and state-backed hackers race to exploit new AI tools. DHS must immediately help states and localities find and fix vulnerabilities before Americans are hit with outages, disruptions, and attacks that could put lives and livelihoods at risk. This is a race between cyber defenders and AI-enabled hackers — and with communities across the country at risk, there is no time to waste.”

]]>

‘It would be insane’ for spy agencies to not have AI model early access, lawmaker says

David DiMolfetta — Tue, 12 May 2026 14:09:00 -0400

Rep. Jim Himes, D-Conn., the top Democrat on the House Intelligence Committee, said Tuesday it would be “insane” for U.S. intelligence agencies to not have early access to advanced artificial intelligence models that could be used for hacking and cyberdefense.

His remarks, delivered on a panel at Politico’s Security Summit, come as the Trump administration is reportedly considering a major AI executive order and debating whether the Commerce Department or intelligence community should oversee evaluations of AI models. They also come as President Donald Trump makes a planned trip to China this week, where he is expected to discuss AI matters with Chinese President Xi Jinping.

“Making sure that, in particular, where our real computational brains are, the National Security Agency, making sure they have access to the most capable hacking tools … it would be insane not to do that, right?” he said.

The NSA, the spy community’s premiere hacking, codebreaking and foreign eavesdropping giant, has been testing Mythos, a major Anthropic model that’s been held back from full public release due to its substantial cyber capabilities, multiple people familiar with the matter said.

On Monday, The Washington Post reported the Trump administration is split over whether to give spy agencies or the Commerce Department dibs at evaluating models. Commerce officials are pushing back against a White House proposal to house an AI evaluation center within the intelligence community, according to the report.

Himes said the Commerce Department should also have a role to play in AI policy.

“Across the government, we should be looking at these capabilities,” he said, adding that “we ought to be cultivating — not damaging — our relationship with the producer of this remarkable new technology,” in a nod to the ongoing legal complaints Anthropic has lodged at the Defense Department, which deemed it a supply chain risk earlier this year after the company said it wouldn’t meet certain Pentagon demands.

Himes said he doesn’t think the legal spat between the DOD and Anthropic has set back the intelligence community in the near term, though “if this drags out, if [Defense Secretary] Pete Hegseth gets a bee in his bonnet about this and just decides to target because his ego is damaged … that will be a massive liability for United States national security.”

Officials are circulating draft policy documents with language clarifying the government’s ability to use private sector tech without outside stipulations, Nextgov/FCW reported last week. It’s not clear if the contracting language is part of a coming executive order or a separate policy initiative.

The ongoing discussions highlight how the Trump administration is closely examining cyber threats brought on by advanced AI models and is looking to take a more hands-on approach toward the AI sector, despite prior laissez-faire positions.

]]>

Lawmakers propose to establish AI guardrails for VA in FY27 funding

Edward Graham — Tue, 12 May 2026 12:21:00 -0400

Proposed amendments to the fiscal year 2027 funding bill for the Department of Veterans Affairs include several measures seeking to limit the use of decisional or unapproved artificial intelligence tools by the agency.

The House Rules Committee is set to hold a hearing on the VA funding proposal and proposed amendments Tuesday afternoon. The FY27 funding package passed out of the House Appropriations Committee last month.

Separate proposals offered by Reps. Paul Gosar, R-Ariz., and James Walkinshaw, D-Va., specifically call for additional oversight of VA’s uses of AI to ensure they are being deployed appropriately. Although both proposed amendments may not make it into the final funding package voted on by the full House, they signal lingering lawmaker unease about the VA’s use of the emerging capabilities to augment department operations.

Gosar’s measure would block VA funding from being used “to make a final determination with respect to the approval or denial of a claim for disability compensation under the laws administered by the Secretary of Veterans Affairs using artificial intelligence.”

VA has used AI and automation to speed up its processing of veterans benefits claims, although some lawmakers and veteran service organizations have expressed concern about officials ceding too much of their decisionmaking authority to the technologies. The agency has stressed that humans always make the final claims decisions and that the AI tools act as more of an automated information retrieval system.

Gosar spokesman Anthony Foti said the congressman’s amendment would “establish a clear safeguard within the VA disability claims process” to ensure that human reviewers always make the final decisions, adding that Gosar “believes it is important to establish appropriate guardrails before these technologies are relied upon for consequential adjudicative decisions.”

Even though VA is already leveraging some AI tools to speed up benefits processing, the agency is looking to further expand its suite of claims-focused technologies. VA’s 2025 AI use case inventory, which was released in January, included 28 instances of the technologies being leveraged for claims processing, with the majority of these examples still listed as being in the pre-deployment phase.

“Disability determinations often involve nuanced medical evidence, individualized circumstances, and credibility assessments that require human judgment and oversight,” Foti said. “The amendment is designed to ensure that AI serves as a support tool — not as the final decision-maker — in matters directly impacting veterans’ healthcare access, financial stability, and earned benefits.”

Walkinshaw’s amendment, meanwhile, looks to address uses of the emerging capabilities that may be operating without agency oversight.

Within 180 days of the funding bill’s enactment, Walkinshaw’s proposal would require VA to submit a report to relevant congressional committees detailing “the use of unapproved artificial intelligence models or artificial intelligence-powered applications, also known as ‘’shadow AI’’, within information technology networks of the Department of Veterans Affairs, and any cybersecurity risks and data exposure vulnerabilities introduced by such use.”

These ‘shadow’ uses of AI often entail employees using technology not approved by the agency to conduct work and that operate outside of traditional restrictions.

Walkinshaw’s office was not able to immediately respond to a request for comment.

]]>

Anthropic and nonprofit partner to streamline benefits administration with AI

Kaitlyn Levinson — Mon, 11 May 2026 16:00:00 -0400

CHICAGO — The civic tech nonprofit Code for America is partnering with artificial intelligence company Anthropic to develop tools aimed at helping caseworkers enhance public benefits administration across the nation.

The organizations are working together to develop an AI-enabled solution to improve the accuracy and timeliness of benefits service delivery under the Supplemental Nutrition Assistance Program, Jana Rhyu, vice president of product at Code for America, announced Friday at a summit hosted by the organization in Chicago last week.

The SNAP Policy Navigator tool is built on federal regulations, state manual selections, official policy directives and other documents to help caseworkers “quickly and accurately get an answer to [a] very specific policy question” when they are working with clients, said Michael Lai, who leads state and local government AI at Anthropic.

The tool leverages Anthropic’s Claude chatbot and is built on a model context protocol to ensure a secure two-way connection between data sources and AI applications, Rhyu said. A caseworker can input a simple, policy-based question, such as how a client’s change in income or a new federal policy could impact their benefits, and the tool outputs an up-to-date response in plain language with cited sources and suggested next steps.

The user “gets clarity on policy, not a decision on overall eligibility. The decision stays with [them],” she said.

The announcement comes as state and local public benefit agencies scramble to comply with rule changes to the federal food assistance program made last July under President Donald Trump’s “Big, Beautiful Bill.” The law subjects participants to expanded work requirements, shifts administrative costs to states based on their SNAP payment error rates, and requires that the Thrifty Food Plan — the model used to calculate the lowest-cost nutritional meal for a family of four — be cost-neutral to changes in food prices.

Since its passage, SNAP participation has declined by more than 3 million people across 36 states as of January, and further reductions are expected once the new rules are fully implemented, according to the Center on Budget and Policy Priorities.

“Policy is constantly changing, and the complexities of policy implementation are immense, which places an even bigger burden on the caseworkers,” Rhyu said.

That’s why it’s critical for resources like the SNAP Policy Navigator tool to help reduce caseworkers’ administrative burden of sifting through and trying to apply intricate policies to individual cases, she said.

Indeed, the complex rules regarding eligibility and exemptions present common barriers to benefits access and having to explain those to residents who depend on the timely and accurate delivery of public assistance to meet their everyday needs only adds to it, Lai said.

Such challenges are exacerbated by funding uncertainty, workforce shortages and increasing caseloads that many states and localities are grappling with across the U.S.

He pointed to one former caseworker who described their job as “an email inbox that's always full, where each one requires care and attention, but you're constantly getting interrupted as you try to work through the never ending inbox of people to help.”

In addition to the SNAP Policy Navigator, Code for America and Anthropic will develop a suite of Claude-based tools to further assist benefit workers with answering policy questions, reviewing eligibility documents and drafting communications to benefit recipients, Code for America leaders said in an announcement last week.

“We know that caseworkers are really overburdened in general, but especially at this moment with HR 1 as well, and so AI shouldn't be used for AI’s sake,” Lai said. “We want it ultimately to be helping in this human way and trying to make benefits administration more efficient, more accurate and more human centered.”

]]>

US tech official calls for ‘transformational’ use of AI in scientific discovery

Edward Graham — Thu, 07 May 2026 16:50:00 -0400

The Trump administration sees greater incorporation of artificial intelligence capabilities into the scientific research space as critical for continued U.S. technology leadership, a White House official said on Thursday.

Speaking at the Special Competitive Studies Project’s AI+ Expo, U.S. Chief Technology Officer Ethan Klein said a major focus of this administration “is having better integration and tie-in across the scientific development piece, all the way through tech development, testing, prototyping and scale up.”

Klein said greater adoption of emerging capabilities like agentic AI — autonomous systems capable of executing specific tasks with minimal human oversight — will have a profound impact on scientific research. A Market Connections survey of more than 200 technology executives across government that was released on Tuesday found that 53% of respondents said their agencies were already exploring uses of agentic AI or were planning pilots of the technology.

“Across a broad swath of applications, but specifically for scientific discovery, I think agentic AI will be transformational,” said Klein, who also serves as an associate director of the White House Office of Science and Technology Policy.

Greater use of these capabilities, he said, would help to expand and enhance data collection and transform the types of experiments that can be conducted by researchers.

“I think that if we're able to actually deploy these agentic AI … agents across those workflows, they're going to see a great amount of scientific efficiency,” Klein added. "And that's incredibly important, because that underpins every one of these technologies that we're looking to develop.”

The Trump administration has already taken some steps to enhance nationwide research efforts by leveraging AI. The largest of these is the Genesis Mission, which was launched in November 2025 and seeks to further harness AI for scientific advancement.

Klein said the initiative will help bring “a bit of that muscle [when it comes to] incorporating that into the workflows that we know are going to bring forth this new era of AI-enabled scientific discovery.”

Thursday’s panel, however, was held amid ongoing concerns about how the Trump administration’s push to scale back government operations through layoffs and reductions in force is impacting research efforts.

Just last month, President Donald Trump dismissed all 22 members of the independent advisory board overseeing the National Science Foundation, which supports nationwide science and engineering research. Critics have said the purge — which comes as NSF still lacks a permanent director — will harm continued U.S. scientific leadership.

Editor’s note: Market Connections is a business division of GovExec, the parent company of Nextgov/FCW.

]]>

Pentagon will ‘never again’ rely on a single AI provider, official says

Alexandra Kelley — Thu, 07 May 2026 14:41:00 -0400

Leadership at the Pentagon reiterated the agency’s commitment to diversifying its artificial intelligence service providers, with Under Secretary of Defense for Research and Engineering Emil Michael taking the stage Thursday at an event in Washington, D.C., to stress that his department is never being “single-threaded with any one model.”

Speaking during the Special Competitive Studies Project’s AI+ Expo event, Michael said that the recent deals between eight leading AI developers and the Department of Defense are both a private sector statement of support for working with the government, as well as a step towards the Pentagon’s goal to diversify its tech stack with different providers.

“We were single-threaded on one vendor, one AI vendor at the Department of War, and to integrate into classified systems is not just putting your software on a public cloud and having it work,” Michael said, referring to his agency’s contract with Anthropic. “These are sophisticated, protective systems that take a lot of work to integrate on, so it wasn't like I could just turn on a few other models that easily. But never again we’ll be single-threaded with any one model.”

Michael continued to say that the new deals with Amazon Web Services, Google, Microsoft, NVIDIA, OpenAI, Reflection, Oracle and SpaceX are “a statement by the biggest tech companies in the world who are involved in the AI space … and have them say, ‘We support the Department of War, we support the U.S. government, and we support the… armed services for all lawful use cases.”

Michael’s comments come in the midst of an ongoing dispute between Anthropic and the Department of Defense following the company’s refusal to have its technology used in operations involving autonomous weaponry and American surveillance.

The fallout of that dispute resulted in the Pentagon designating Anthropic a supply chain risk and the White House ordering agencies to begin removing the company's products from their tech stacks. A judge put a hold on those actions in late March pending ongoing litigation over the government’s actions.

The release of Anthropic’s advanced cybersecurity-focused model, Mythos Preview, changed the discussion. Access to Mythos and its advanced capabilities for detecting cybersecurity flaws is tantalizing for the U.S. government, prompting internal drafts of policy plans that would enable some agencies to use Anthropic’s cutting-edge model.

Michael said that the advent of Mythos signals the forthcoming evolution of cyber-capable AI models.

“The Mythos moment is really a cyber moment, and it's: ‘How is the U.S. government going to deal with cyber?’” Michael said.

Major tech companies are responding to Michael’s drive to diversify the Pentagon’s vendor portfolio. Rand Waldron, the vice president of the Global Government Sector for Oracle Cloud Infrastructure, told Nextgov/FCW that Defense officials are asking cloud service providers like Oracle to prioritize interconnectedness in the effort to avoid vendor lock-in.

“From what I can see, the Department of War has some very savvy people who … don't want to go all in on one [model] because then six months later, they may need to go all in on another,” Waldron said.

He explained that there will likely be models that are more finely-tuned to particular use cases, such as code generation, data analytics, supply chain management or targeting in warfighter operations. One model from a single provider may not effectively serve each of these workflows.

“I don't believe that all those different use cases will end up being the exact same model at any given time,” Waldron said.

The Pentagon’s desire to expand the service offerings available for its workforce has precedent. Waldron said that DOD and the intelligence community have laid the foundation for a flexible approach to AI services acquisition, citing the creation of the Commercial Cloud Enterprise and Joint Warfighting Cloud Capability contracting vehicles as the blueprints for future contracting structure.

“It's not like they're trying to replace Anthropic with another model provider,” Waldron said. “They want to replace Anthropic with four model providers.”

]]>

FDA launches updated AI and consolidated data platform

Alexandra Kelley — Wed, 06 May 2026 15:02:00 -0400

The Food and Drug Administration announced on Wednesday that it has launched an updated version of its internal artificial intelligence tool, called Elsa, that is being integrated with a consolidated data platform to enable faster search capabilities.

Elsa 4.0 now features expanded access to disparate agency information within a new consolidated data platform, called the Harmonized AI & Lifecycle Operations for Data. By merging HALO and Elsa, staff are able to query data and build workflows with less manual updating.

Elsa 4.0 runs on Google Cloud Platform and is built within a FedRAMP High secure designation. It isn’t trained on input data or data submitted by regulated industry, in order to safeguard sensitive information handled by FDA personnel.

“Elsa’s new capabilities once again position FDA as a leader in deploying AI tools that empower staff,” FDA Commissioner Marty Makary said in a press release. “Removing tedious burdens for staff enables them to focus more on science and makes their work streams more efficient and enjoyable. We have some of the best scientists in the world and we need to take good care of them.”

The upgrades to Elsa 4.0 include custom agentic AI, document generation, data analysis and visualization functions, voice-to-text diction, web search features and conversion of scanned documents into searchable text, among other capabilities.

The FDA initially deployed Elsa in June 2025 with a core goal of hastening the scientific review and evaluation process. The agency has also recently turned to AI to run a pilot for real-time clinical drug trials that will pull a direct data feed on the trial for FDA staff to expedite analyses and clearances of devices, drugs and medications.

The FDA has been on a broad IT modernization journey in recent months, consolidating duplicative systems and software licenses, using the savings it has achieved to reinvest in the scientific community, new technologies and to onboard as many as 3,000 new scientists. The agency has also worked to scale up employee use of generative AI from just 1% in early 2025 to over 80% today.

]]>

Commerce AI center will evaluate Google Deepmind, Microsoft and xAI models

Alexandra Kelley — Tue, 05 May 2026 17:37:00 -0400

The Center for Artificial Intelligence Standards and Innovation will be conducting testing on leading AI models from Google Deepmind, Microsoft and xAI to evaluate their security prior to deployment, the Commerce Department announced Tuesday.

CAISI, housed within the National Institute of Standards and Technology, will oversee the testing as well as best practices development related to commercial AI systems. The models will be tested in classified environments.

The agreements between Google Deepmind, Microsoft and xAI and Commerce build off of earlier voluntary agreements, and were renegotiated to support the Trump administration’s AI Action plan.

“Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” said CAISI Director Chris Fall. “These expanded industry collaborations help us scale our work in the public interest at a critical moment.”

CAISI’s evaluations will look at the national security-related risks and capabilities of each model. This effort hinges on information sharing between CAISI and model developers, and CAISI will study models that have reduced or removed safeguards to better understand their unmitigated capabilities.

Prior to evaluating U.S.-based AI models, CAISI recently examined Chinese model DeepSeek, concluding it underperformed in several areas like accuracy, security and cost efficiency.

The announcement follows recent reports that the administration is considering an executive order that would create government protocols to test AI models prior to market deployment. The news was first reported by The New York Times on Monday and confirmed to Nextgov/FCW on Tuesday.

Among industry groups, initial reactions to the agreements have been supportive. Business Software Alliance Senior Vice President of Global Policy Aaron Cooper said that CAISI brings the necessary expertise to work with private sector partners to evaluate frontier models for safety and national security risks.

“Today’s announcement reinforces CAISI’s role as the right institutional home within government for advancing evaluation and measurement science and convening AI companies and stakeholders on a voluntary basis around responsible practices,” Cooper said in a statement. “BSA has highlighted why frontier model evaluation should be led at the federal level, reflecting the national security implications at stake; a strong role for CAISI can also help further global collaboration and alignment on safety and security.”

]]>

Agency leader says AI is helping resource-strained workforce identify more fraud

Sean Michael Newhouse — Tue, 05 May 2026 16:00:00 -0400

An anti-fraud official from the Centers for Medicare and Medicaid Services said that artificial intelligence is enhancing her workforce’s capability to spot scams.

“There's a lot of fraud in the healthcare sector. The estimates on the conservative side are about $100 billion and, depending on who else you talk to, you can easily double or triple that with different calculations,” said Jeneen Iwugo — the acting director for the CMS Center for Program Integrity — at the UiPath Public Sector Summit on Tuesday. “I have a modest budget of $1 billion, so the size of the problem is much bigger than the budget I'm given to find it.”

Specifically, Iwugo explained that AI tools are helping CPI’s roughly 500 employees better identify fraud across the four to five million claims they review every day.

“My team uses AI to comb through those claims and that data and figure out where the risk is the greatest,” she said. “I cannot investigate everything that looks weird, so I have to restratify my work to make sure that I am auditing and reviewing those instances where we have the biggest risk for something fraudulent happening.”

Iwugo emphasized that the Trump administration’s prioritization of combating fraud has given her office more flexibility.

“The longer leash I get, the more I'm able to push the needle with using AI, getting into the agentic AI space, the more of that $100 billion I'll be able to recapture,” she said.

Going forward, Iwugo stated that CMS is piloting programs in which AI doesn’t just review potentially fraudulent claims but also recommends to employees what the possible penalties could be.

“Once I get there, I will be able to take off and capture a lot more of the fraud that I know exists, that I'm able to detect, but that I'm just watching because I can't move fast enough,” she said.

Iwugo also touted that CPI in fiscal 2024 saved Medicare an estimated $26.3 billion, which is a return on investment of $14.6 for every $1 invested and an improvement from fiscal 2023 when the agency projected that it recovered $14.9 billion.

Agencies reported there were 3,611 individual AI use cases in 2025, which is more than double from 2024. The Veterans Affairs Department, as one example, is using the technology to speed up processing of veterans’ benefits claims, but congressional Democrats argued that it is worsening error rates.

]]>

Trump admin floats policy language limiting contractor say on agency uses of technology

Alexandra Kelley and David DiMolfetta — Tue, 05 May 2026 14:12:00 -0400

The federal government is circulating draft policy documents that contain language clarifying the government’s ability to use private sector technology without outside stipulations for how they do so, two sources familiar with their development told Nextgov/FCW.

While it remains unclear if the language being passed between various government agencies — namely the Department of Defense and components of the Trump administration — will manifest into an executive order or finalized policy, that language centers on ensuring the government has control over how its acquired technology products are used.

One source familiar with the ongoing development told Nextgov/FCW that the goal of the language is to clarify that “it is for that democratically elected government to determine what is a lawful and appropriate use of a particular technology, not solely a company.”

The White House is mulling an executive order that would create a working group for AI models before they are deployed, according to a person familiar with the matter. The New York Times first reported the administration’s consideration of the order. It’s not clear if the contracting language is a separate initiative or would be a provision embedded into a forthcoming directive.

Other language featured in the draft documents examines how the government can manage emerging cybersecurity threats posed by AI models like Anthropic’s Mythos Preview and OpenAI’s GPT 5.5, according to the same source and another person familiar with the matter.

The discussions and draft documents highlight how the Trump administration is looking to take a more hands-on approach on the AI sector, despite prior policy positions that signaled a more permissive environment for the evolving technology.

“There is likely going to be another wave of AI government statements,” the first source said.

When asked to confirm the existence of these documents, a White House official told Nextgov/FCW that “any policy announcement will come directly from the President. Discussion about potential executive orders or policy directives are pure speculation.” The Department of Defense referred questions to the White House.

The Trump administration’s efforts to refine the government’s rights when licensing private sector AI models and systems follow a dispute between Anthropic and the Department of Defense over using the company’s AI products in autonomous weaponry and domestic surveillance.

The Pentagon designated Anthropic a supply chain risk, and federal agencies were subsequently required to offload the company’s products from federal workloads. Some lawmakers took issue with the perceived retaliation on behalf of the administration, and Rep. Sam Liccardo, D-Calif., attempted to amend the Defense Production Act to prevent government blacklisting in March.

The administration has telegraphed some of its wants for the relationship between vendors and industries since that debacle, with Emil Michael, the undersecretary of defense for research and engineering, saying on a March episode of the All-In podcast that “all lawful use seems like a good thing” to benchmark against.

Anthropic model capabilities piqued government officials' interest, however, when the company announced the release of its new high-powered Mythos Preview model and associated Project Glasswing for select companies to test in their digital networks.

Leadership at the Pentagon has not been wholly opposed to guardrails on tech use in defense and warfighter operations. Michael told CNBC on May 1 that the Pentagon wants guardrails “in some ways,” but maintained that these guardrails have to align with the government’s needs.

“When they deploy on our networks, they’re deploying models that are tuned for national security purposes,” Michael said. “And that's why the partnership with the executive team and the management is so important, because things are evolving in the threat landscape. And whatever guardrails, whatever principles they want to develop against, has to be consistent with our values, our mandate, our restrictions, even, and that’s where the guardrails come in.”

]]>

Survey: More than half of federal agencies now planning agentic AI pilots

Frank Konkel — Tue, 05 May 2026 10:00:00 -0400

Agentic AI has entered the chat for federal agencies.

According to a March survey of more than 200 technology executives across government, more than half (53%) said their agencies are exploring agentic AI or actively planning pilots of the technology. Another 15% are currently implementing agentic AI systems or have completely done so already, compared to 6% who said they were not yet considering agentic AI.

However, even as agencies race toward agentic AI, respondents identified potential barriers to adoption in the forms of inadequate oversight policies and disparity between the perceived necessity of governance frameworks and their implementation. The findings were released today by Market Connections on behalf of ServiceNow.

“This research confirms what we're hearing from agencies every day — the appetite for agentic AI is real, but oversight hasn't kept pace," said Mike Hurt, Global Vice President of U.S. Public Sector at ServiceNow. "Seventy-seven percent of federal leaders say oversight frameworks are essential, yet fewer than a third have actually implemented them. Agencies that build accountability into their AI workflows from the start, not as an afterthought, will be the ones delivering strong results for citizens.”

The findings come after a major uptick in AI use across the federal government in 2025 despite a significant decrease in the total number of federal employees.

In April, the Office of Management and Budget unveiled its 2025 Federal Agency Artificial Intelligence Use Case Inventory, which indicated AI use more than doubled across federal agencies from 2024. In total, agencies reported more than 3,000 AI use cases, with significant jumps in AI use at NASA and the departments of Health and Human Services, Veterans Affairs, Justice and Energy.

Agentic AI is generally defined as autonomous systems capable of pursuing complex goals and reasoning, with the ability to take independent actions across software systems with minimal human oversight. Those agentic capabilities to perform some tasks without human intervention have made it an attractive option, and it has been touted by some of President Donald Trump’s top tech officials as a key way to do more with less.

But per the findings, not every agency is ready for agentic AI, even as a growing number of companies offer agentic solutions. Only 20% of respondents said their agencies have defined policies for pre-deployment testing or generic agentic AI use, and only 8% have a defined framework for incident response. Even fewer (6%) have a framework for third-party or vendor governance.

"Agentic AI has definitely entered the chat. Over half of those we surveyed are currently in the planning stages or have actively launched a pilot effort,” said Aaron Heffron, president of Insights and Research at GovExec. “The main question remains, however, if the current infrastructure, both human and technical, is up to the task.”

Findings also indicated some agencies struggle moving agentic AI pilots from the sandbox to production environments. Those challenges move beyond policy and into data readiness.

“One of the very big questions that you have to ask is, ‘How am I getting my data ready for AI consumption? That governance piece becomes critical [to] making sure that your data within your organization and your AI are working together,” advised one unnamed IT director in his response.

ServiceNow Federal Chief Technology Officer Jon Alboum said one way to address the data problem is bringing data and workflows together in a single environment.

“The current environment of fragmented, siloed systems and disconnected workflows only increases complexity and hinders adoption,” Alboum said. “To move forward, AI adoption should focus on bringing everything together in an AI control tower so that policies can be applied, controls enforced, and results delivered efficiently across the organization.”

Research findings indicated that while governance frameworks tended to lack maturity, consistent oversight was a near-universal requirement. Almost 90% of respondents said they required logging and audit trails for all actions, and more than 80% requiring automated policy checks and guardrails.

The findings further point to a strong correlation between the demand for human oversight and criticality of data. For national security, critical infrastructure and emergency response data, 79% of respondents said their agencies mandated “human-in-the-loop” oversight, with approval needed for every action performed by AI. For high-risk data, like benefits claims or agency financial data, 78% of respondents said their agency requires formal human approval before high-risk actions are taken by AI, but not every action. Conversely, more than 90% of those surveyed said they favored reduced direct involvement for low and moderate risk data, requiring only periodic check-ins.

To ensure accountability for agentic AI solutions, 84% of respondents said their agencies had documented escalation policies, while 78% had structured post-incident review processes. Fewer than half (44%) said their agencies included liability or responsibility clauses for AI vendors in contracts, and fewer than one-third (29%) had documented “kill switch” procedures.

“Federal leaders say they want human control over high-risk AI — but less than a third have a kill switch to enforce it, leaving a dangerous gap between intent and capability where trust is won or lost,” the report states. “Agencies must act now to define intervention triggers, ensure data readiness, and unify oversight into a single platform — or risk losing control as systems scale. When failures happen, they cannot be crises; they must be contained, repeatable workflows. Built-in accountability is no longer optional — it is the prerequisite for any agency serious about deploying agentic AI at mission speed.”

Editor’s note: Market Connections is a business division of GovExec, the parent company of Nextgov/FCW.

]]>

Pentagon makes agreements with 8 companies to add AI to classified networks

Alexandra Kelley — Fri, 01 May 2026 11:49:00 -0400

The Pentagon announced new agreements with eight leading artificial intelligence developers on Friday, with those companies set to support classified workflows in military operations.

SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, Oracle and Amazon Web Services have all been selected to deploy their AI products in the Department of Defense’s Impact Level 6 and Impact Level 7 network environments, with a specific focus on streamlining data synthesis, improving warfighter decision-making and elevating situational understanding and awareness.

Notably, the press release said these new contracts are part of the Pentagon’s ongoing efforts to “build an architecture that prevents AI vendor lock and ensures long-term flexibility for the Joint Force” by making a diverse range of AI tools available for Defense employees.

“Together, the War Department and these strategic partners share the conviction that American leadership in AI is indispensable to national security,” the release said. “This leadership depends on a thriving domestic ecosystem of capable model developers that enable the full and effective use of their capabilities in support of Department missions. As mandated by President [Donald] Trump and Secretary [Pete] Hegseth, the Department will continue to envelop our warfighters with advanced AI to meet the unprecedented emerging threats of tomorrow and to strengthen our Arsenal of Freedom.”

The new AI tools will be available via GenAI.mil, the Pentagon’s central AI platform. In late April, Google rolled out its Gemini 3.1 Pro model on the platform for various defense use cases.

The announcement follows tensions that exploded between another AI company, Anthropic, and the Pentagon in late February after the company refused to allow its products to be used for autonomous weapons and surveillance of Americans. The Pentagon subsequently designated Anthropic a supply chain risk and Trump ordered federal agencies to begin offloading use of its products, though a judge has issued an injunction on those actions.

Editor's note: This article has been updated to reflect the addition of Oracle in the Department of Defense press release.

]]>