Nextgov/FCW - All Content

USPTO looks to AI agents to ease examiner workloads

Edward Graham — Tue, 20 Oct 2026 18:25:00 -0400

Although agentic artificial intelligence systems are poised to help the U.S. Patent and Trademark Office significantly reduce the time it takes to review applications by streamlining rote tasks, a top official with the agency said the capabilities also have another added benefit — empowering skilled examiners to rely more on human-to-human interactions rather than the typical stream of impersonal correspondence.

Barry Schindler, USPTO’s acting commissioner for patents, said the agency currently has a diverse staff of patent reviewers, including PhD biochemists who review relevant patent applications. Rather than bogging these personnel down with tedious tasks during the three-step pre-exam, examination and post-exam patent lifecycle process, Schindler said USPTO has turned, in part, to a new agentic tool — called Compass — to help.

Examiners reviewing man-made or altered DNA sequences, for instance, first have to see if the submitted sequences are correct before determining whether or not they can be patented.

Schindler said these reviewers frequently have to engage in a “back and forth, back and forth, back and forth” dialogue with applicants, which is largely conducted through written correspondence. Now, Schindler said Compass “does what's called a formality check” to speed along the process.

“So then what happens now? I've got a PhD biochemist who had been hired to do what they were supposed to do,” Schindler said during the Google Public Sector Summit on Tuesday.

With a current backlog of 700,000 applications for review, and a staff of around 9,000 patent examiners, USPTO needs to turn to AI agents for help, he said, adding that further adoption of these tools can help promote more consistency in the review process.

“You as an applicant may get an examiner who has 20 years of experience, and at the same time you may get an examiner who has two years,” he said. “We put our examiners online almost immediately; we have to. We have this backlog. So what we're looking to do with this whole [AI] and [agentic agents thing] is how do we start getting consistency? So that examiner who has two years, how do we impart that knowledge that's similar?”

Providing patent examiners with a consistent review process and giving them AI tools to help navigate often cumbersome tasks, Schindler said, means skilled employees can more fully leverage their talents.

“The thing that is missing — and we don't have time to do it — is the human interaction,” Schindler said, noting that patent examinations are often “a negotiation” between applicant and examiner.

“If part of that iterative process is, instead of the examiner just putting paper — they do paper — back and forth, can we get the examiner on the phone and start having this human interaction? And examiners will say, ‘I love that, but I don't have time because I'm doing this [other stuff],’” he said.

Rather than taking sometimes five years for applicants to receive a patent, Schindler said “we can shorten that to one year, because you can pick up the phone and have conversations” by using agents to help shoulder some of the other workload.

Adoption of these tools, however, doesn’t need to be large. Schindler said even small uses of these capabilities are beneficial and noted that some research studies have shown that too much reliance on AI tools can impair cognitive skills.

“If somebody starts relying on agents and [they] don't understand the fundamentals and don't continue to learn the fundamentals, they can get dumb,” he said. “And so I have that issue that I have to think about: [using] agents, but how does the agent assist and not take over?”

Focusing on the underlying problem, rather than a one-size solution, Schindler added, is a necessity when it comes to maximizing the effectiveness of new capabilities like those offered by agents.

USPTO has previously touted some of the other AI tools it’s rolled out to help improve its internal operations, including an enterprise AI tool named Scout — a chatbot assistant used by personnel to help analyze, draft, research and summarize content.

]]>

NASA’s self-driving Mars rover sets an off-world mileage record

John Breeden II — Tue, 20 Oct 2026 18:21:00 -0400

The classic lunar rovers that astronauts drove during the Apollo missions have become some of the most recognizable vehicles ever built. They raced across the moon in grainy television broadcasts during the 1970s and have since appeared in countless documentaries, movies and museum displays. But despite their fame, none of them ever made it more than about 22 miles.

On Oct. 11, NASA’s Perseverance rover quietly surpassed them all, along with every other vehicle ever operated on another world. Its odometer reached 28.10 miles, or 45.22 kilometers, narrowly exceeding the 28.06-mile record set by NASA’s Opportunity rover.

NASA marked the occasion with an animated map showing the winding route Perseverance has followed since landing in Jezero Crater in February 2021. During that journey, it crossed an ancient lakebed, climbed through a fossilized river delta and reached the crater’s rim.

Now, it might have endurance, but it won’t be breaking any speed records. The rover needed about five and a half years to set its distance record. And the rover it just passed, Opportunity, took nearly 15 years to establish the previous benchmark.

Perseverance, which started its mission in 2020, has a lot of technical improvements compared with Opportunity, which began exploring Mars in 2004. It’s also a larger vehicle built for endurance. But it’s not just a matter of Perseverance having better wheels, more power and a higher top speed. One of the key reasons for Perseverance’s ability to go the distance is how much driving authority NASA has gradually transferred from human operators on Earth to software operating on Mars.

More than 90% of the distance traveled by Perseverance was covered using its AutoNav autonomous navigation system. By comparison, NASA says that Opportunity and its twin rover, Spirit, used an earlier form of the AutoNav system for about 5% of their travels.

The AutoNav system is important because a command from Earth can take more than 20 minutes to reach Mars, depending on where the planets are in their respective orbits around the sun. Waiting for a response can stretch a simple exchange considerably longer. That is a long time for a rover to sit at a crossroads while operators decide whether it should go left or right around an obstacle. Ideally, AutoNav allows Earthbound operators to assign a destination and let the rover determine the safest way to get there.

The earliest forms of the navigation program showed good proof of concept, but were generally not advanced enough to trust with longer journeys, especially over the incredibly challenging Martian terrain. For example, the first NASA rover to operate on Mars, the microwave oven-sized Sojourner, could only evaluate about 20 different points in the terrain when deciding where it could safely move, enough to spot obvious pitfalls but not some of the more hidden challenges or trouble spots lying in wait along its chosen path.

By contrast, the upgraded AutoNav system inside Perseverance processes 600,000 points of terrain every single time it looks at the ground ahead. That in turn allows it to create a three-dimensional map, identify rocks, steep slopes and sand traps, then select a safe route between destinations chosen by human operators. According to Vandi Verma, a senior engineer, rover planner and driver at NASA’s Jet Propulsion Laboratory, the process is called “thinking while driving” because the rover is processing its route while its wheels are turning.

This new capability gives Perseverance the ability to travel up to 1,351 feet during a single Martian day. That might not impress anyone accustomed to Earth’s highway speeds, but it represents a major improvement for a robotic vehicle moving through an uncharted desert without roads, GPS or immediate human assistance.

NASA is also experimenting with using generative AI to remove another potential bottleneck from the rover’s travels. AutoNav can decide how to avoid a rock or sand trap once Perseverance is moving, but human rover planners have traditionally needed to study a lot of orbital imagery and terrain data to select the waypoints that define its broader route.

During two demonstration drives in December 2025, NASA instead used vision-language models to analyze high-resolution orbital images and terrain-slope data. Working with the same information used by human planners, the AI identified bedrock, outcrops, boulder fields and sand ripples, then generated a continuous route and the waypoints needed to follow it.

Perseverance traveled 689 feet during the first AI-planned drive and 807 feet during the second. NASA engineers did not blindly accept the results. Before transmitting the commands to Mars, they ran the proposed routes through a digital twin of the rover and verified more than 500,000 telemetry variables to ensure the instructions were compatible with its software.

“The fundamental elements of generative AI are showing a lot of promise in streamlining the pillars of autonomous navigation for off-planet driving: perception, localization, and planning and control,” Verma said.

That test did not turn Perseverance loose to select its own scientific destinations. Humans still decided where the rover needed to go, while generative AI helped plot the route and AutoNav handled obstacles encountered along the way. But those technologies are steadily removing the pauses, manual planning and localization checks that once limited how much ground a rover could cover.

NASA recently extended Perseverance’s mission through approximately the end of 2028, and all of its mobility systems remain fully operational. The rover is now traveling toward an area called Idubi, where scientists hope to find evidence of ancient hydrothermal systems and learn whether those environments might once have supported life.

“The rover that broke this record did so by making its own decisions about the path,” said Mark Maimone, deputy team chief for robotic operations on Perseverance. “Given the autonomy to just go, Perseverance is a better driver than the one that landed, and we’re not done. Our newest software teaches Perseverance to steer less and roll more, and to check its own position against orbital maps mid-drive.”

That means 28.10 miles probably won’t remain the record for very long. But the more significant achievement is how Perseverance reached it. NASA has steadily moved navigation decisions from operators on Earth to increasingly capable systems on Mars, giving the rover more time to explore and less time waiting for someone back home to tell it which way to turn.

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

]]>

State Department orders ‘super intelligence’ wording across all official communications

David DiMolfetta and Jory Heckman — Tue, 20 Oct 2026 13:28:00 -0400

The State Department has instructed personnel to replace “artificial intelligence” with “Super Intelligence” in official communications and rename its AI-related programs and systems, bringing President Donald Trump’s terminology change into the daily work of U.S. diplomacy, according to two people familiar with the matter and contents of an unclassified cable obtained by Government Executive and Nextgov/FCW.

The cable — dated Monday and addressed department-wide — tells staff to use “Super Intelligence,” or “SI,” in new letters, emails, public messages, websites, reports, remarks and policy documents, subject to legal exceptions. Existing material should be updated when staff next revise it.

“All official titles and the names of Department governance bodies, systems, platforms, inventories, and programs must use SI in place of AI,” the cable states.

The instructions implement Trump’s Sept. 29 executive order, which directed executive agencies to adopt the new terminology across official communications. The cable shows how that broad instruction is being translated into specific requirements for diplomats and the department’s internal operations.

The order says the wording should reflect the advancing capabilities and promise of American technology. For now, however, it defines “Super Intelligence” as the same technologies and systems covered by the existing statutory definition of AI. In computing science, super intelligence is defined differently than the generative artificial intelligence technologies used today.

In recent weeks, administration officials had previously discussed the terminology change with certain State bureaus, people familiar with the matter said. But the cable appears to be the first organization-wide directive outlining where the new language must be used. The Associated Press previously reported such communications with State’s international organizations bureau.

The transition, while expected for some time, is significant because it would change how U.S. diplomats describe the fast-evolving technology in policy documents and discussions with overseas counterparts.

Foreign organizations retain their own names in internal writings. The cable specifically tells employees not to change references to China’s World Artificial Intelligence Cooperation Organization, even as the department replaces the term in its own materials.

The notice also spells out several exceptions to the replacement rule. Employees must preserve language required by law and existing legal citations, including quotations from statutes, regulations, treaties and other legal instruments. Contracts and grants are also excluded, pending further guidance.

Previously published documents, cables and archived web pages should not be revised or reissued solely to change the terminology, it adds. The instructions also preserve the published titles of existing presidential actions, programs and organizations, such as America’s AI Action Plan, unless the department provides further direction.

One program has already changed names. The cable says the American AI Exports Program is now the American SI Exports Program, citing an earlier State Department cable. The program’s public website already uses that name to describe its work helping American companies sell technology overseas.

For staff handling international agreements or departmental rules, the cable directs questions about terminology to the appropriate policy or legal office.

The vocabulary change is already appearing in senior officials’ public remarks. National Cyber Director Sean Cairncross repeatedly referred to AI as “super intelligence” and “SI” during an Oct. 1 discussion on AI safety, following Trump’s order.

Office of Personnel Management CIO Adam Starr also used “SI” in remarks last week, describing efforts to help agencies adopt the technology. Starr said OPM had launched a SuperIntelligence Governance and Adoption Program, or SI GAP, to assist other agencies.

]]>

Democrats warn tech firms of election security ‘backsliding’ ahead of midterms

David DiMolfetta — Tue, 20 Oct 2026 11:07:00 -0400

Top congressional Democrats are pressing major technology companies to strengthen their election safeguards ahead of the midterms, warning that inconsistent enforcement and inaccurate AI-generated voting information could leave Americans more exposed to manipulation or interfere with their ability to cast ballots.

In a letter to the heads of 25 companies — including Alphabet, Meta, Microsoft, OpenAI and TikTok — the lawmakers said some firms are retreating from commitments to protect elections, even as their products make it easier for foreign adversaries and domestic actors to reach large audiences with deceptive material.

“Unfortunately, there are concerning signs of backsliding on some of your commitments to protecting elections from misuse of your products,” they wrote. “These signs are particularly alarming because they come at a time when the capabilities of malign actors are growing in scale and sophistication and our elections are under attack from multiple directions.”

The letter was signed by Sens. Mark Warner, D-Va., and Alex Padilla, D-Calif., and Reps. Jim Himes, D-Conn., and Joe Morelle, D-N.Y.

Warner and Himes are the top Democrats on their chambers’ intelligence committees, while Padilla and Morelle hold corresponding roles on the Senate Rules and House Administration committees, which oversee election matters.

Their appeal comes amid what several stakeholders see as weakening cooperation between federal agencies, technology companies and outside researchers tracking foreign influence operations, as Nextgov/FCW reported last week.

Nextgov/FCW has requested comment from TechNet and the Computer & Communications Industry Association, trade groups that count several recipients of the letter as members, including Google, Meta, OpenAI and Anthropic.

For the last two years, the Trump administration has dismantled multiple hubs focused on foreign influence and election disinformation, largely on censorship grounds. Federal agencies have contended that election security support continues despite these changes.

“Public reporting indicates that Russia continues to undermine confidence in U.S. elections through disinformation campaigns executed directly on many of your platforms; though such efforts clearly violate your terms of service, some of your companies have not quickly or consistently removed such content,” the lawmakers wrote, likely referring to recent Alethea research detailing the Russian influence efforts.

Russia is likely to pursue more overt influence operations in the coming midterms, while China is likely to play a longer game in the information space in order to attain broader geopolitical goals, a current U.S. official told Nextgov/FCW last week. Both nations are expected to make greater use of advanced AI tools for their efforts.

The lawmakers’ requests focus, in part, on information that can affect whether someone successfully votes, including registration deadlines, polling locations, identification requirements and mail-in ballot rules. Companies should distinguish deceptive claims about those procedures from political speech, and remove “demonstrably false” information once they become aware of it, the letter says.

They also want platforms to give election officials dedicated tools to report false information, use independent fact-checkers alongside internal teams and automated systems, and notify users when content they previously engaged with is later determined to be false or misleading.

As voters turn to chatbots for information about candidates and voting procedures, incorrect or outdated answers can also create problems, they said. “Equally alarming is that some of your AI chatbots have an unacceptably high error rate when responding to queries about election and voting processes, subjecting users to confusing or inaccurate answers that could result in voter disenfranchisement,” the lawmakers wrote.

A September study by the Institute for Strategic Dialogue documented chatbot errors that could misdirect voters. In two responses, Meta’s Muse Spark model incorrectly identified Nov. 4 as the date of this year’s midterm elections, a day after the actual election.

Some firms have publicly outlined election preparation plans, including Meta and TikTok. Last week, X also disclosed a takedown of thousands of accounts believed to have been carrying out foreign malign influence operations.

The lawmakers also challenged reductions in trust and safety staffing and urged companies to give independent researchers access to products and data needed to assess whether safeguards work.

“Refusing to grow and expand your election integrity teams — and even shrinking these teams — does not support your stated commitments to countering harmful election content and suggests to bad actors that your policies are not enforceable,” they wrote.

]]>

Google Public Sector unveils Gemini for government services

Christian Robles — Tue, 20 Oct 2026 06:00:00 -0400

Google Public Sector on Tuesday unveiled Gemini for government services — an add-on that gives federal, state and local governments access to specialized skills, artificial intelligence agents and integrations with frequently used software.

The suite of tools can help government officials interpret dense regulations, redact documents requested through the Freedom of Information Act, translate jargony documents into simple language and review contracting bids, in addition to completing other tasks, according to a company press release.

Select agencies, including the federal Health and Human Services Department and the State of Missouri, have been testing Gemini for government services over the past few weeks, Chris Hein, field CTO at Google Public Sector, told Nextgov/FCW. The company is accepting additional early adopters of Gemini for government services to field feedback before general availability is expected later this year, he added. There is no target date for general availability.

“Gemini for government services provides a secure environment for agencies to automate workflows, allowing civil servants to focus on delivering the responsive, high-quality services that citizens expect,” said Thomas Kurian, CEO of Google Cloud, in the release.

Gemini for government services includes pre-built, specialized skills that serve as a set of rules instructing Gemini to execute a task without going off script, Hein told Nextgov/FCW. Government officials can refer to a skill in the Gemini prompt bar that would help analyze regulations, audit expenses and conduct other activities, according to a video demo.

Officials can also access third-party AI agents from other companies, such as Accenture Federal Services and Deloitte, to further assist government work, the release says.

Google Public Sector will “prescreen” the AI agents, which agency IT leadership, not individual users, can enable through a behind-the-scenes portal, Hein explained to Nextgov/FCW. Some agents are already available but most are not yet, he added.

Gemini for government services includes secure connections to technologies officials already use, such as Google Workspace, Microsoft SharePoint, Outlook 365, Salesforce and ServiceNow, the release says. That connector feature extends to federal regulations by digital services company BLEN, allowing officials to search the Electronic Code of Federal Regulations, regulations.gov and the Federal Register. The feature includes a connection to Oracle enterprise apps, according to the release.

When Gemini for government services achieves general availability, agency IT leadership will be able to choose which connections they want to enable for which set of users, Hein said.

In the demo, a compliance analyst for a state chief information security officer used Gemini for government services to analyze federal cybersecurity regulations with the pre-built policy and statute assistant skill. After the analysis, the user asked Gemini to draft a ServiceNow incident ticket to remediate contracts for compliance with the federal cybersecurity rules. The user could edit the incident ticket before sending it.

Federal agencies have access to Gemini for $0.47 per agency per year through Nov. 15 under a recently extended OneGov deal.

]]>

Permitting reform ‘single most important thing we could do for national security,’ senator says

Edward Graham — Mon, 19 Oct 2026 18:05:00 -0400

Permitting reform should be a top priority when it comes to ensuring that America maintains its technology dominance over global adversaries, according to a U.S. senator.

During an event at the Hudson Institute on Friday, Sen. Dave McCormick, R-Pa., said the United States needs to out-build China — particularly when it comes to artificial intelligence capabilities — as a bulwark against Beijing’s world power aspirations.

But AI advancements are leading to increasing electrical needs and the construction of more data centers — areas where, McCormick said, China’s policies could allow it to outpace the U.S.

“I can assure you, China's permitting challenges are not ours,” the senator said. “If we take five-and-a-half to six years for an energy project, they take less than six months. And so their energy production is really something we have to compete on, and right now we can't do that without permitting.”

With Congress on recess until after the midterm elections next month, McCormick said one thing lawmakers should quickly prioritize in the lame duck session is legislation to speed up this process.

“I think the single most important thing we could do for national security is permitting reform, because the permitting reform bill would streamline our capacity to put energy production in place,” he said.

Lawmakers in both chambers of Congress have introduced a variety of proposals over the past year to speed up the permitting process. McCormick has championed several reform-focused measures as well, including co-sponsoring a bipartisan bill introduced by Sen. Mark Kelly, D-Ariz., in February that looks to modernize the permitting process by turning to cloud-based technologies.

One of the more notable reform bills was introduced last month by Sens. Martin Heinrich, D-N.M., Mike Lee, R-Utah, Shelley Moore Capito, R-W.Va., Sheldon Whitehouse, D-R.I. Their proposal, which McCormick alluded to, would overhaul the permitting process by, in part, imposing deadlines on environmental reviews, limiting the period for legal challenges and creating new digital tracking tools.

“If you think energy production is a necessity for our economic well-being but also our national security, because energy feeds the compute power, compute feeds the AI capability … then permitting is probably the biggest thing we could do,” McCormick said.

Although data center construction has become a hot-button issue, with lawmakers on both sides of the aisle introducing various bills to limit their impact on surrounding communities, McCormick voiced his support for the infrastructure development.

“To have AI leadership, you have to have compute,” he said. “To have compute, you have to have energy. To have this, you need data centers.”

He blamed some of the negative coverage of data center development, in part, on “Beijing's efforts to poison public opinion around this issue” — something that President Donald Trump has also echoed, and something that congressional Republicans have also been investigating.

But McCormick said data center construction is something that should ultimately be decided by the local citizenry, requiring “a compact between the community and the hyperscalers that are building the data centers.”

]]>

HHS issues seven nonconformity letters in crackdown on health data blocking

Christian Robles — Mon, 19 Oct 2026 18:03:00 -0400

Santa Fe, N.M. — The Health and Human Services Department has issued seven letters of potential nonconformity to companies that may have illegally blocked access to health data, an agency official said.

The companies receiving letters have been able to remediate allegations of restricted data access and “for the most part, folks have said, ‘you know what, you're right, we shouldn't have been doing this, and we're going to change our behavior,’” Thomas Keane, the national coordinator for health information technology, said Thursday at the ADAPT conference. He did not disclose which companies received the letters.

Still, the previously unreported number of letters of potential nonconformity is the Trump administration’s latest step to ensure health data moves. Last September, HHS said it would take an “active enforcement stance” against health care entities that illegally block patients, providers and other stakeholders from gaining access to information.

The 21st Century Cures Act, signed into law in late 2016, prohibits entities from engaging in that practice, which is known as “information blocking.” Later regulations from the first Trump and Biden administrations spelled out penalties for entities that are found to have committed information blocking, including a $1 million civil penalty per violation.

But the Trump administration has yet to announce concrete enforcement of information blocking rules. Keane told ADAPT attendees his office is working with the HHS Office of Inspector General, Justice Department and Federal Trade Commission to enforce the regulations.

At the same time, Keane said his office wants to work with health IT companies and other regulated entities in ensuring they understand information blocking rules and are willing to change data sharing practices.

“If not, we take an enforcement posture,” he said.

While HHS seeks to give patients access to their medical records through information blocking enforcement, the Centers for Medicare and Medicaid Services is holding “troubleshooting call[s]” to help patients get their data, agency Deputy Administrator and Administrator of the U.S. DOGE Service Amy Gleason said.

The calls, of which there have been two thus far, bring together patients, electronic health record companies and medical providers to figure out how to get health data flowing. CMS has gotten patient data to move from the two calls, Gleason said.

]]>

Report: How to make government websites AI friendly

Kaitlyn Levinson — Mon, 19 Oct 2026 16:54:00 -0400

Artificial intelligence is altering how constituents interact with their governments and services, as the technology is increasingly woven into benefits administration, public transportation and, now, website navigation, according to a new report. This presents an opportunity for governments to revise their public websites to ensure they remain a relevant source among artificial intelligence-enabled searches.

“As more residents turn to AI assistants instead of visiting a government website directly, their ability to find the right answer increasingly depends on if that assistant can actually reach — and correctly interpret — government content,” a new report released last week from Georgetown University’s Beeck Center for Social Impact and Innovation reads.

The report aims to help government staff responsible for technical and content decisions regarding government websites — such as IT offices, chief information officers or web design teams — prepare their content to be accessible for AI tools used to identify and relay such information, according to the report.

Those tools are known as AI crawlers, or automated programs varying in scope and purpose that sweep websites to collect information, which are increasingly requiring agencies to consider how they design website access, structure and content accuracy, Andrew Merluzzi, artificial intelligence and incubation fellow at Georgetown University’s Beeck Center and report author, told Route Fifty.

Generally, this “is not a super difficult thing for governments to fix, but they have to know they have this problem in the first place in order to go fix it,” he said, adding that “the main goal of this [report] was to try to help governments make the information transparent that they're already trying to make transparent.”

One step government can take to improve AI crawler access is to assess their current website security configurations. Many websites include settings that block “malicious” bots, which could “inadvertently block a legitimate AI assistant trying to answer a person’s question,” the guidance reads.

However, agencies must also balance AI crawler access with its potential impact on a website’s hosting capacity, due to increased traffic, according to the report.

Government can consider a three-tier system to help determine how they approach granting access to certain AI crawlers through their website settings, such as a robots.txt file, Merluzzi said.

The first tier of AI crawlers refers to tools from “established AI providers,” which the report recommends that a government allow to access its website when their identities can be verified. Such crawlers or crawler setting controls include OpenAI’s GPTbot, Anthropic’s Claude and Google’s Googlebot, among others.

In fact, “you don’t have to care about every possible [large language model] on the planet accessing your website — what you need to care about are the major tools that people are using every single day,” Merluzzi said.

AI tools under the second tier may be developed from established providers but could have “narrower relevance” for resident use or have documented compliance concerns, according to the report. Before governments design their websites to allow such AI crawlers, staff can monitor and evaluate specific tools on a case-by-case basis.

For bots that do not match a verified bot signature, which fall under the report’s third tier, government can implement a graduated response system of website access. For instance, staff can configure websites with unique responses to different bots, such as logging a potential bot that appears low risk for further review or automatically blocking traffic from a source highly likely to be a bot, Merluzzi explained.

The report also points to website structure as a tool for building access for AI crawlers.

Governments are spending considerable resources and time to improve the usability and interactivity of their websites with features like drop-down menus and clickable dashboards, but plainness and simplicity could be key with AI crawlers, Merluzzi said.

“The implication here now for AI is just [to] publish a lot of that information as plain text, and that’s all you need to do so that the LLMs can then find a webpage … and read it and relay it back to the user who’s asking a question,” he said.

In addition to more interactive webpages, government staff can publish critical information — such as deadlines for benefit program applications — as web page text, HTML tables or PDFs with extractable text and data, according to the report. Otherwise, scanned documents or screenshots containing relevant data for users may be overlooked by AI crawlers.

Staff should also retest a website’s interaction with AI crawlers following major website redesigns or backend changes to a content management system, as such alterations “can affect whether AI crawlers and retrieval systems can access important information, even when the visible content appears unchanged,” according to the report.

The accuracy of information on a website is another critical pillar to AI crawler preparation, Merluzzi said.

Government should designate a specific person or team, separate from managing the website’s security against AI crawlers, to focus on content accuracy and timeliness, according to the report.

These individuals can own the process for updating and correcting information across government websites, prioritizing data “consequence” rather than “volume,” the report reads.

“A rarely-viewed page with a wrong emergency number matters more than a frequently-viewed page with a minor outdated statistic,” according to the report, which also recommends testing if an AI assistant pulls the correct and updated website data when queried.

Unlike other “technology problems in state and local governments for which the solutions are not easy [because] they’re bureaucratic or likely to take many years,” Merluzzi said, reconfiguring websites to be more searchable by AI crawlers “is one where if you put a few people on this problem for a week or two, you could make really substantial progress without a ton of expertise.”

]]>

Congress is already paying for loan systems. It should pay for one.

Robert Shea and Doug Criscitello — Mon, 19 Oct 2026 15:45:00 -0400

The federal government holds more than $5 trillion in direct loans and loan guarantees. Those loans finance farms, homes, small businesses, students and exports through more than 100 credit programs at about 20 agencies. Each agency also buys, operates and replaces its own loan systems, and none of them share a common lexicon or data standard. The government can’t readily assess its credit risks in real time, compare performance across programs or determine whether an applicant denied at one agency was approved at another.

Sens. Marsha Blackburn, R-Tenn., and Maggie Hassan, D-N.H., with Reps. Brad Finstad, R-Minn., and Raja Krishnamoorthi, D-Ill., introduced legislation to correct that. The Federal Loan Systems Modernization Act would authorize a shared lending platform using commercially available technology, with GSA responsible for acquisition and OMB for coordination. Under the bill, eligibility rules, underwriting criteria, interest rates and servicing policy would remain with the programs. But the back office, meaning intake, underwriting workflow, servicing, document management and fraud detection, would be consolidated and modernized.

In an analysis for the Center for USA Lending, we estimated how the Congressional Budget Office would likely score the bill. We think the net cost would be close to zero. Enactment requires an appropriation in the low tens of millions of dollars over five years for planning, data standards and oversight. After that, agencies pay a cost recovery fee from the same accounts they already use to run their own systems. Because that fee moves money from one federal account to another, CBO would likely treat it as an offsetting collection with no net effect on the deficit. That estimate is correct under the rules, but it omits most of what the bill would save.

On operating costs, we think the government could save between $250 million and $300 million a year once the platform matures, or $1 billion to $1.5 billion over ten years, net of transition costs and after a three-year migration. There are precedents. GAO found that data center consolidation produced $6.24 billion in savings and cost avoidance between 2012 and 2020, and payroll consolidation — the model this bill follows — projected about $1 billion over ten years for a much simpler function. CBO will not count any of it, though, because the bill has no direct impact on any specific appropriation, and an agency that spends less of its funding on one activity may spend the difference on another.

Much of the fraud in federal lending occurs at origination, when an applicant misrepresents their identity, income, eligibility or the existence of the business being financed. SBA's Inspector General estimates that more than $200 billion in pandemic disaster and Paycheck Protection Program funds went to potentially fraudulent actors. When SBA later employed better screening at the front of the process, it blocked 21.3 million duplicate, ineligible and fraudulent applications across its four largest pandemic programs. A consolidated, modern loan management system would give all participating government loan programs the benefit of the same rigorous fraud controls.

Today, though, agencies are replacing their systems one at a time. Ginnie Mae has a modernization contract valued at about $127 million. FHA is spending about $92 million on Catalyst. VA is recompeting its home loan guaranty systems, which cost $328 million over four years. USDA announced its own effort in June. None of these systems are designed to exchange data with the others. Congress is funding this work either way, and the question is whether it buys numerous incompatible systems or one, or perhaps a few, that every program can use.

Two administrative steps would help. OMB could require agencies to report operational and avoided modernization savings in their budget submissions. OMB and Treasury could require every migrating program to screen applicants against Treasury's Do Not Pay service, which helped agencies prevent, detect and recover $11.7 billion in fiscal 2025, and to report losses. That would give CBO the evidence it needs to score fraud savings in the future.

Farmers waiting on loans before planting, first-generation homebuyers and small businesses without access to bank credit would get their money faster and with less exposure to fraud. Congress should pass the bill.

Doug Criscitello, former CFO at SBA and HUD, is the executive director of the Center for USA Lending, a nonpartisan organization which advances modern design and delivery of federal credit programs. Robert Shea, former Associate Director at OMB, is CEO of GovNavigators.

]]>

Senator warns hundreds of federal web addresses may lack key spying protection

David DiMolfetta — Mon, 19 Oct 2026 09:00:00 -0400

Sen. Ron Wyden, D-Ore., is urging the Trump administration to close an encryption gap that could let foreign adversaries track visits to sensitive government websites, including the FBI’s tip portal and a veterans crisis line, according to a letter sent to the Cybersecurity and Infrastructure Security Agency and the Office of Management and Budget on Monday and first seen by Nextgov/FCW.

Wyden asked CISA and OMB to require that agencies implement a web security standard that hides the name of the website someone is visiting. The protocol, known as Encrypted Client Hello, protects the website name when a user’s browser connects to a site and has been supported by major browsers and some web infrastructure providers since 2023.

Wyden’s office identified 350 federal web addresses that lack the encryption specification. They include sites that facilitate intelligence community job applications, immigration services, health information and U.S. diplomatic websites around the world.

“This visibility could lead to malicious eavesdroppers identifying whistleblowers, profiling cleared federal personnel, tracking Americans seeking crisis support, and mapping interactions between agencies,” Wyden wrote to CISA acting director Nick Andersen and deputy federal chief information officer Thomas Flagg.

He also asked OMB, working with CISA and the Office of the National Cyber Director, to require that cloud and network providers support the technology to qualify for federal contracts. He requested responses by Nov. 20.

Nextgov/FCW has requested comments from CISA and OMB.

Federal websites generally encrypt the information people read or submit, though hackers can still learn which sites they visit by monitoring their internet traffic.

The letter singles out two services whose visitors could be particularly vulnerable.

The Department of Veterans Affairs’ Veterans Crisis Line platform uses Amazon Web Services, which does not currently support ECH for the site, according to the letter. Wyden warned that the exposure could help foreign adversaries identify veterans seeking mental health assistance. Additionally, the FBI’s tips.fbi.gov site uses Cloudflare, a provider that supports ECH, but the portal has not enabled it, the senator added, arguing the current arrangement could put people contacting the bureau at risk of identification.

The possible FBI exposure also comes after a prolific criminal hacker group recently stole troves of sensitive employee information tied to the bureau’s jobs portal.

The list of sites without the protocol also includes hotline.oig.dhs.gov, the DHS inspector general’s complaint hotline; civilrights.justice.gov, the Justice Department’s civil rights reporting portal; and apply.intelligencecareers.gov, an intelligence community recruitment site. Federal HIV information sites, including hivinfo.nih.gov and clinicalinfo.hiv.gov, are also listed.

To craft the list, Wyden’s staffers tested federal web addresses using third-party network providers to see if ECH successfully hid which specific websites were being visited, a spokesperson told Nextgov/FCW. The protection only appeared to work on sites relying on Cloudflare’s free service tier, the spokesperson added.

The letter does not establish that adversaries have tracked visitors to any of the listed addresses.

Wyden argued the government can close these gaps using technology already supported by major browsers and deployed across millions of websites. For agencies whose providers support ECH, enabling it can be as simple as changing a setting, he said.

“Modern federal zero trust cybersecurity practices rely on protective software installed directly on agency devices, so enabling [Encrypted Client Hello] would strictly improve the cybersecurity of federal networks,” Wyden wrote.

The effort follows his September call for NSA to update its advice about virtual private networks. A Congressional Research Service analysis requested by his office at the time warned that foreign intelligence services with broad visibility into internet traffic could potentially connect VPN users to websites by comparing traffic patterns.

]]>

VA pushes external providers to adopt tech streamlining veteran specialist appointments

Christian Robles — Mon, 19 Oct 2026 08:00:00 -0400

The Department of Veterans Affairs is encouraging outside providers to enable the External Provider Scheduling, or EPS, system which lets VA staff directly book specialist appointments for veterans with those providers.

The EPS platform is meant to save VA staff the hassle of booking specialist appointments with non-VA providers through faxes and phone calls. All VA facilities implemented the EPS system in late 2025, and the system has 27,000 participating non-VA providers covering 78 specialties, according to a March 2026 press release.

“We’re really cooking with grease and being able to schedule veterans much faster,” Howard Manuel, VA deputy senior advisor, told Nextgov/FCW in a recent interview.

Manuel will tout the scheduling platform during an Open@Epic conference panel on Thursday alongside representatives of Deaconess Health System — which became an early adopter of the EPS system in February — as part of the VA’s effort to push more non-VA providers to enable the technology.

Because of that adoption, Marion VA Medical Center staff can schedule specialty appointments at Deaconess Illinois, a part of the broader health system.

Andrew Beck, system director of business development for Deaconess Health System, told Nextgov/FCW that 688 veterans have been successfully booked to Deaconess Illinois through the EPS system. At one point this year, veteran appointment times at Deaconess Illinois averaged under five minutes, which is significantly lower than the time it would take to book them through phone calls, he added.

Deaconess Illinois has seen a nearly 30% increase in year-over-year VA patient encounters, which William Davis, its regional president, says is the result of improving veterans’ access to care.

Veterans who have received care through the Deaconess Health System have their medical information shared back to the VA through the EpicCare Link system, Beck said.

Deaconess Health System is now planning to give VA staff visibility into radiologist appointments in Illinois and expand the EPS system to other locations beyond the state, he noted.

Deaconess Health System has enabled the scheduling system by integrating its Epic electronic health records with the WellHive technology that underpins the EPS program. But providers using non-Epic electronic health records can share specialist appointment information through the EPS system, according to a VA fact sheet.

WellHive is working to allow VA staff to easily see specialist appointments for VA and non-VA providers on its platform, Chris Faraji, the health technology company’s president, told Nextgov/FCW. VA awarded WellHive a contract for the EPS system in 2023.

Faraji added that the company is working to enable the EPS program for health systems around the country. Beyond Deaconess Health System, WellHive is already working with the Medical University of South Carolina, University of Arkansas Medical School, Allegheny Health Network and West Virginia University on the EPS system, he said.

“The momentum is there. It's really starting to pick up,” Faraji said.

At the beginning of the second Trump administration, Rep. Tom Barrett, R-Mich., pressed VA to rapidly expand the adoption of the EPS system and blamed the Biden administration for the technology’s slow rollout. Secretary Doug Collins later told lawmakers in June 2025 that the VA is doing “everything we possibly can” to boost the scheduling system.

VA’s focus is now to “open up the door to other health systems to join us in this partnership,” Manuel told Nextgov/FCW.

]]>

Tech bills of the week: Cutting federal funding for Flock; Making data centers shoulder their power costs; and more

Edward Graham — Fri, 16 Oct 2026 17:25:00 -0400

Stopping federal funds from going to AI-powered surveillance cameras

Rep. Rashida Tlaib, D-Mich., introduced legislation on Wednesday that would prohibit federal funds from being used to purchase or operate any automated mass surveillance systems, including AI-powered cameras provided by Flock.

In a press release, the congresswoman’s office said her proposal — the Get the Flock Out of Our Communities Act — “is the only bill in Congress designed to completely prohibit the use of automated mass surveillance technology, including biometric surveillance systems and facial recognition technology.”

Sen. Josh Hawley, R-Mo., introduced a separate measure late last month that would impose data guardrails on the use of the surveillance systems. Rep. Pat Harrigan, R-N.C., also introduced a measure on Thursday that would “restrict the usage of automated license plate reader system by governmental entities,” although a public version of that bill’s text is not yet available.

“Flock’s technology isn’t making us safer,” Tlaib said in a statement. “Instead it has been used to stalk people, mis-identify and arrest innocent people, surveil protestors, and wrongfully record our neighbors. My bill would prevent our government on all levels from selling us out to Big Tech companies that trample our civil rights for profit.”

Making data centers pay for their power usage

Rep. Eugene Vindman, D-Va., introduced a measure on Wednesday that seeks to keep communities from having to foot the bill for data centers’ power expenses by requiring large and medium-sized data centers “to pay the full incremental costs associated with their electricity use.”

The bill would apply to data centers and high-density computing facilities that have electricity demands of at least 20 megawatts.

“As data centers and artificial intelligence drive growing demand for electricity, we need to make sure the companies benefiting from that growth pay their fair share,” Vindman said in a statement.

Removing tax breaks for data centers

Sen. Bernie Moreno, R-Ohio, also introduced legislation on Thursday that would prohibit data centers from receiving federal and state incentives for their development or operations.

Details of Moreno’s measure, known as the Recapturing Excessive Corporate Largesse through Accountability and Incentive Measures — or RECLAIM — Act, were first reported by Politico.

The outlet said the senator’s bill includes a provision stating that “States and political subdivisions thereof should promptly repeal or reform any tax incentive programs that are specific to data centers.”

Mandating disclosure of AI use in political ads

Rep. Yvette Clarke, D-N.Y., reintroduced legislation on Thursday that would require political advertisements to include a disclaimer if they include any artificial intelligence-generated photos or video.

Clarke first rolled out the measure in July 2023 and told Nextgov/FCW at the time that “for those who want to be deceptive or to harm their opponents, they can easily do some real damage with AI-generated advertising.”

The congresswoman’s bill calls for relevant ads “to include a statement within the contents of the advertisements if generative AI was used to generate any image or video footage in the advertisements.”

In a press release, Clarke said that “if AI content can manipulate and deceive people on a large scale, it can be a serious threat to our democracy and have devastating consequences for our national and election security.”

Creating a new federal platform to combat elder scams

Reps. Marie Gluesenkamp Perez, D-Wash., and Tom Barrett, R-Mich., rolled out legislation on Thursday that looks to crack down on fraud and scams targeting elderly Americans by creating a new website and 24/7 reporting hotline.

The measure looks to address the often fragmented fraud reporting system across government by creating a streamlined reporting and assistance process. The lawmakers said their bill would create a reporting platform similar to the one that the Department of Veterans Affairs created in 2024 to combat scams targeting retired servicemembers. Barrett chairs the House Veterans’ Affairs Technology Modernization Subcommittee.

“Families need resources to recognize and quickly report suspected fraud,” Barrett said in a statement. “By establishing a 24/7, one-stop shop for addressing elder fraud, this bipartisan bill will protect vulnerable seniors and empower law enforcement to crack down on these criminals.”

]]>

The Fed's 35-year open data bet is a model for the AI era

Nick Hart — Fri, 16 Oct 2026 16:27:00 -0400

Ask an economist, a reporter or a legislative staffer where a number came from and the answer often sounds like a colleague's name: FRED. Federal Reserve Governor Christopher Waller said the same is true around the world. Mention the St. Louis Fed overseas, he told a conference audience on Oct. 1, and people answer with one word: FRED.

I first met FRED long ago, as a novice user working for the leadership of the Missouri House. Even with little experience, I could quickly pull the information decisionmakers needed. Years later, as an economic researcher, I relied on FRED for far more sophisticated analytics. Few public tools serve both of those users well, and that range is a big part of what makes FRED a public good.

This month I was back in Missouri at the Federal Reserve Bank of St. Louis, moderating the closing panel of that conference on where FRED, the bank's Federal Reserve Economic Data service, goes next. Waller opened the day by calling FRED "the greatest public good" the Fed has ever created. After a day with the people who build and use it, I think he has a case.

These days FRED is making a lot of new friends and about half of them aren't human. For government executives, how FRED got here and how it is preparing for those new visitors, is the lesson worth studying.

A 35-year model

FRED's roots go back to 1961, when St. Louis Fed research director Homer Jones began mailing a typed weekly digest of economic data to policymakers, academics and journalists. In 1991, the bank put that data online through a computer bulletin board that users reached by dial-up modem. FRED became a website in 1995 with 865 data series and about 6,000 users a week.

Today FRED hosts more than 850,000 data series from governments and other sources around the world. It reached 18 million unique users in 2025, according to Waller.

The format changed with every new technology. The value proposition never did: one consistent, noncommercial place to find data from many producers, reproduced accurately and credited to the agencies that create it. That is what open data advocates have asked of the government for years, and what Congress wrote into the Foundations for Evidence-Based Policymaking Act of 2018 through the OPEN Government Data Act. FRED was delivering on that vision decades before the law passed.

Open data is not free

FRED did not grow from 6,000 weekly users to 18 million annual users by accident. Three ingredients made it work, and every agency leader should recognize them.

Sustained investment. Each leap, from mailed digests to dial-up to the web, required someone to fund staff and infrastructure ahead of demand. That pressure is now accelerating. Waller reported that FRED traffic is growing about 150% a year, driven mostly by AI agents and bots. Serving that load reliably costs real money, and agencies whose data draws similar attention will face the same bill.

Listening to users. The St. Louis Fed filled a room with the people who produce and depend on its data: officials from the U.S. Bureau of Economic Analysis and the U.S. Census Bureau, economists from Amazon, FedEx, Ramp, Indeed and Visa, journalists from The Economist, The Wall Street Journal and Bloomberg TV, and researchers from Kalshi and the World Bank. Keith Taylor, who leads the bank's research information services, closed the day by inviting participants to submit interest in the new FRED Advisory Board to keep those users involved. He was also candid that his team is better at saying "trust us" than at showing the work behind that claim. Leaders willing to say that out loud are usually the ones who fix it.

Leadership. Waller credited the bank's leaders for keeping their eyes on what FRED could become, and he championed FRED himself as the bank's director of research. Leadership also showed up in a decision he announced at the conference: users looking for certain data published by the Federal Reserve Board of Governors will soon be routed to FRED. Consolidating access on a proven, trusted platform is a choice leaders have to make on purpose.

The next generation of open data

The most striking number of the day: about half of visits to FRED now come from AI agents retrieving data, Waller said. The data suggests we’re seeing a seismic shift in how people consume information that originates on a portal like FRED. In catering to their users, public agencies responsible for making data open and usable have a different primary audience now. A website designed for human eyes doesn't automatically work for software that reads, summarizes and repackages data for someone else.

The St. Louis Fed's answer is the FRED MCP Connector, built on the Model Context Protocol, an open standard for linking AI systems to outside data. It lets people find and retrieve FRED data through the AI agent of their choice rather than a single tool chosen for them. FRED staff are also building guidance to help agents work with economic data accurately, and linking FRED to ALFRED — its archive of data as first reported — and to FRASER — the Fed's library of historical documents — so AI answers can show how numbers were revised and the context behind them.

Being AI-forward here is not about chasing hype. Waller was candid that AI can hallucinate, misread data and confuse correlation with causation, and that FRED's managers can't always tell whether agents credit the original source. The connector is a bet that the best defense is to make the trusted path the easiest path. On the closing panel I moderated, we kept returning to a simple test: is access actually improving, meaning better-informed users, not just more traffic? When half the visitors are machines, that question gets harder to answer and more important to ask.

Others in the room are learning the same lesson. Ara Kharazian of Ramp, whose AI Index went live on FRED the day before the conference, described first publishing the company's AI Index as an embedded chart, then discovering that AI tools couldn't read it. Ramp rebuilt the page so agents could. Aidan Smith of the Census Bureau described statistical agencies working to make their websites ready for AI so agents pick up methodology and limitations along with the numbers. And FedEx's Arvind Rajagopalan argued that generative AI raises the stakes on the unglamorous fundamentals: metadata, quality and freshness.

What federal leaders should do now

Every agency that publishes data is about to face FRED's situation, if it hasn't already. Three steps would help.

First, treat AI agents as an audience. Test whether common AI tools can find your data, read it correctly and credit your agency as the source. If they can't, fix the publishing, not the user.

Second, fund the fundamentals. Documentation, metadata, machine-readable formats and Chief Data Officers with real resources are not overhead. They are what keeps data trustworthy when no human is checking the output.

Third, build feedback loops that include the developers and AI companies whose tools now sit between agencies and the public. FRED's conference is one model. Agencies don't need a ballroom to start; they need a habit of asking.

The Data Foundation and its partners are developing a National Data Strategy that treats public data as national infrastructure, starting from the reality that generative AI makes data both more valuable and more sensitive. FRED's experience belongs at the center of that conversation.

In the early days of FRED, the team mailed typed pages. Today a student, a small-business owner or a congressional staffer can just ask an AI tool a question and get an answer drawn from FRED. The delivery method has changed again and again. But the mission has not: trusted data, open to everyone, serving the American people is key. And that’s why our leaders must keep investing resources in a national data ecosystem that serves us all.

Nick Hart is president and CEO of the Data Foundation, an independent, nonpartisan think tank in Washington, D.C. He moderated the "Accessible for All" panel at the Federal Reserve Bank of St. Louis FREDcon conference on Oct. 1, 2026.

]]>

Utah AI sandbox launches new health care pilots — with question marks

Christian Robles — Fri, 16 Oct 2026 15:50:00 -0400

SANTA FE, N.M. — Utah announced three new pilots last week to test artificial intelligence capabilities that can refill chronic condition-related prescriptions, write initial acne prescriptions and offer virtual physical therapy, but a regulator acknowledged the state still has key questions to address as the model scales nationwide.

Utah regulators have “evidentiary standards, [and we are] working on governance guidance. We're working on a liability bill to begin building the statutory framework around” its AI sandbox, Zach Boyd, director of Utah’s Office of Artificial Intelligence Policy — or OAIP — said at the ADAPT conference in Santa Fe, New Mexico on Thursday.

Through the AI sandbox, Utah has allowed companies to test new technologies in exchange for regulatory relief and state oversight. Utah started AI pilots in 2024 with a mental health support app, but the sandbox did not garner national attention until this year, as the state allowed AI to refill certain prescriptions with clinician oversight through a pilot with health technology company Doctronic.

The Utah Medical Licensing Board fiercely opposed that effort, calling for an immediate suspension of it in an April letter to the Utah Department of Commerce and criticizing the state for not consulting them.

“We hit a nerve,” Boyd said.

The Utah Medical Licensing Board has since formed a working group that meets with state regulators once a week, he explained, adding that the détente came from six months of discussions and updates to the AI pilot.

In an interview with Nextgov/FCW, Boyd noted that Utah removed some drugs from the list of prescriptions AI can refill following scrutiny from the Utah Medical Licensing Board. The list of 190 drugs dropped headache reliever Butalbital and antiarrhythmic medication Flecainide, according to OAIP.

But Utah regulators are still mulling over how the state should split legal liability between AI and the human clinicians who oversee them if patients are harmed.

Regulators have interpreted the law to mean that the AI sandbox does not have authority to change liability rules, but they have pulled several short-term levers to address the question, Boyd explained. OAIP has required Doctronic and AI pilot companies that have engaged with the sandbox since then to carry special liability insurance and not waive liability.

However, “in the long run, we think [insurance is] less sustainable, and we need a way to hold the vendor directly liable for the actions of their automated system,” Boyd said, adding that regulators expect the default is that physicians will bear too much liability.

Boyd hopes the state will unveil a liability bill this month or next month during Utah’s interim legislative session. The bill will kick off a “multi-year process” to refine liability, he said.

Another big question facing Utah’s sandbox is whether the Food and Drug Administration will regulate pilot AI technology. The FDA generally requires software that treats or diagnoses patients to undergo pre-market authorization, which some lawyers told STAT could include the pilot AI technology like Doctronic’s AI prescription refiller.

The FDA has so far been silent on the Utah experiment.

Boyd declined to speak for the FDA but said he hopes to work with federal regulators on creating a unified regulatory pathway for AI.

“It’s really appealing to people to have kind of a vet once, deployed everywhere, aggregate oversight across jurisdictions” model, Boyd told ADAPT attendees.

“A model like that feeds nicely into the federal government and the kind of oversight that they’ll also want to exercise,” he added.

Utah has tapped six independent evaluators to assess AI companies’ technical claims and share results with the state. They include the Coalition for Health AI, Clarion AI Partners, Glacis Technologies, mpathic AI, Vega Health and Stanford Clinical Excellence Research Center.

Much like with liability, there are question marks about the independent evaluators.

Utah spelled out the general expectations for health care software piloted through the state’s AI sandbox. The technical document says those technologies should account for managing AI input uncertainty, clinical context, patients’ emotional intelligence and other factors throughout its pilot.

“Half of [the document] is probably wrong, but it's definitely, it's got to be, about the most mature thing that exists,” Boyd told ADAPT attendees.

With those expectations, future AI pilots will have to contract with an independent evaluator, Boyd said.

The data evaluators will assess will depend on the specific AI pilot’s intended application, patient population, outcomes and specifications, Mark Sendak, CEO and co-founder of Vega Health, told Nextgov/FCW.

The startup would work with the state and the pilot company on crafting the appropriate data analysis while bringing in external partners with the necessary expertise to assess specific pilots, he said.

But because AI models often change over time and will expand to serve new patient populations, Sendak expects independent evaluators to constantly repeat assessments and share that data with the state.

Utah has also signed master agreements with health systems Intermountain Health and University of Utah Health, allowing them to pitch AI pilots that the state can approve.

“One of the underrated advantages of a sandbox is it helps the regulator learn so much,” Boyd told ADAPT attendees.

]]>

Former Transportation CIO most proud of creating ‘momentum for change’

Frank Konkel — Fri, 16 Oct 2026 08:00:00 -0400

When Pavan Pidugu was sworn in to his role as the Transportation Department’s chief digital and information officer in February 2025, the agency operated more than a dozen separate grant systems, few of its rank-and-file employees used artificial intelligence in their daily jobs and the agency didn’t have an IT strategy for the modern age.

By the time Pidugu left his role on Sept. 4, he’d engineered eye-opening reversals in the above statistics and more, executing a vision he first laid out in Transportation’s comprehensive “1DOT IT plan” that will dictate IT operations over the next four years.

Consider the post-Pidugu statistics: Transportation is consolidating to a single grant system through its GrantLens program, and more than 90% of its 50,000 employees regularly use AI for everything from engaging with internal chatbots and agentic systems to coding software.

During the same 19-month period, Pidugu led a Transportation-wide transition to Google Workspace, which included a massive email migration to Gmail, achieving both improved collaboration tools for employees and substantial cost savings over its previous legacy environment. The department is pacing toward migrating 90% of its data to the cloud by the year’s end and has adopted a unified cloud management framework called T-cloud.

Beyond any singular achievement, though, Pidugu said he’s most proud of taking decisive action and creating the conditions for sustained momentum in tech modernization.

“We did some things really well. We did some things great. There are some opportunities we could have done better, but we can only make mistakes when we’re doing something,” Pidugu said in a sit-down interview with Nextgov/FCW, shortly after he sent a final farewell email to staff.

Pidugu took Transportation’s top tech job with more than four years of experience at the department, He had also worked as the chief technology officer at the Federal Motor Carrier Safety Administration, after holding positions at retail giants Target and Walmart.

Unlike many new CIOs installed in the early days of President Donald Trump’s second administration, Pidugu came in with prior experience in the department, a list of goals and a clear view of changes necessary at Transportation on day one.

“I knew what was required to be done because I was in DOT for four years before becoming the CIO, so for me, it was all about, how can I drive change?” Pidugu said. “And I think probably my biggest accomplishment, I would say, was to create that momentum for change. My biggest hope is that [Transportation] keeps that momentum going.”

An early vision for IT

Early in his tenure, Pidugu faced the same headwinds most of his new CIO peers faced: workforce reductions in line with the administration’s new policies and uncertainty over which staff might take the deferred resignation offer. The uncertainty “slowed down” some things, Pidugu said, but he wasted little time implementing a vision for IT based on processes, people and products.

Pidugu’s authority as CIO increased following a memo from Transportation Secretary Sean Duffy banning all shadow IT, disallowing any technology not under the CIO’s purview.

“I want the OCIO within Transportation to be a technology shop where we build technology. That’s the culture we want to create in-house. We're building applications. We're building products, along with being everything else,” Pidugu said during an interview televised on C-Span in June 2025.

In September 2025, Transportation released its 1DOT IT plan, a “transformative” strategy that set forth four strategic goals — eliminate technical debt, become customer-centric, build obsessive products and grow talent — and divided the department’s IT teams into two complimentary domains: “digital factories that deliver mission-focused products and centralized shared services that provide governance, common platforms, standards and enablement.”

“What are the challenges that we face today? While we are in this major transformational journey, there are lots of gaps, and our people, our processes need to change,” Pidugu said in a keynote speech in May at the GITEC 2026 conference. “We're simplifying the way we work. We're simplifying the assets that we have, the technology applications we have, and we are unifying our processes to make the best use of technology that's available today.”

A pivot to Google

Last September, Transportation inked a five-year contract worth up to $89 million to move from Microsoft’s collaboration suite to Google’s cloud-based productivity and collaboration tools, including Gmail, Docs and Meet, and incorporate Google’s popular AI tools, Gemini and NotebookLM. Department staff have fully migrated to Gmail, with a complete migration for component agency FAA expected in the coming months.

The deal was a direct result of discounts on Google and other tech providers’ tools provided through the General Services Administration’s OneGov program, and it has resulted in cost savings of 70%, Pidugu said. Proudly, Transportation achieved discounts on top of those negotiated through OneGov, he added.

Today, all 50,000-plus employees of Transportation and FAA have access to Google Gemini and NotebookLM, and Pidugu said more than 90% of the department’s staff regularly use them.

Pidugu said he’s been impressed by the “recurring incremental efficiency” agency users have gotten through Gemini, noting that it’s common for users to create workflows that shave hours off their workday. Those are hours, he said, that compound weekly and can be better spent on more meaningful tasks.

Transportation’s more advanced AI users have access to more AI models available through GSA’s USAi platform, including Grok.

One grant system to rule them all

Each year, Transportation provides billions of dollars in grant funding through various discretionary and formula programs. Historically, however, it’s a complicated process that involves anywhere from 10 to 14 different systems. The department’s consolidation of those systems is “in progress,” with a goal of condensing them to a single, smart system by mid-next year, according to Pidugu.

Moving to the cloud

Transportation is on pace to move about 90% of its data into the cloud by the end of this calendar year, according to Pidugu, with 40% of the migration already complete. That will free up dollars through efficiency improvements and in expenses to manage internal data centers.

“At the end of the day, [the migration] will avoid a lot of costs,” Pidugu said. “The hardware in our data centers are getting ready to be depreciated, so we have to invest that money. I’m sure some of that money will pay for cloud storage, but we will be freeing up our resources.”

Noteworthy, however, is the T-Cloud framework the agency has leveraged to manage the migration of its applications to the four leading cloud providers: Oracle, Amazon Web Services, Google Cloud and Microsoft. Transportation aims to be cloud agnostic and use each provider where it makes sense.

“Instead of having four different groups managing the cloud four different ways, T-cloud allows us to have a unified process, practices, tools, everything — and we can also automate a lot of it,” Pidugu said.

The Treasury Department, Pidugu said, implemented essentially the same system through a contract it held with a vendor the agency also happened to have an existing relationship with. So rather than bid out a contract to build it, why not save time and money by using the same source code and adjusting as needed?

The idea stemmed from a conversation Pidugu had with Treasury CIO Sam Corcos, then-federal CIO Greg Barbaccia and Eric Sidle, formerly CIO at the Department of Housing and Urban Development and now at the Interior Department. Pidugu received the source code in a .Zip file.

“Treasury had done the majority of the work, so we were able to borrow their source code — a copy of it — and we implemented it, and we integrated with the tools that we in DOT were using or want to use,” Pidugu said.

Ultimately, the shrewd move — a form of “fed-sourcing,” a term the foursome used — significantly reduced the time and money required to build T-Cloud at Transportation, he added.

Meet DOTi

Transportation’s latest AI agent, DOTi, is agentic AI built using ServiceNow’s EmployeeWorks software, formerly known as the MoveWorks platform. The conversational AI software is available to all agency employees except those in FAA, and it was integrated at launch with Transportation’s IT service management workflow.

Users can engage DOTi through Google Chat. A user experiencing an issue with a laptop charger could, for example, ask DOTi what to do. DOTi will “run through the knowledge” and give the user possible solutions, including whether to order a new charger. DOTi can then execute the task, too.

“DOTi may come back and say, ‘Hey, looks like you need to order a new battery or charger, do you want to order one?’ If you say, ‘Yes,’ it creates your system workflow, and you don’t have to call in to the help desk” Pidugu said.

Across town, the State Department is also using the EmployeeWorks platform, within its HR system. That system allows State employees to inquire about leave balance, schedule days off and more. In the future, Transportation may take similar action and add additional workflows to DOTi’s purview to further reduce call volumes and help-desk backlog.

“The ultimate goal is to leverage these implementations and minimize the call volumes and save,” Pidugu said. “A lot of organizations are talking about AI, but they only look at creating a chatbot or creating generative AI usage. This is true agentic AI. You are using the AI agent to do 10, 15 different jobs — jobs that would take time.”

Beyond vibe-coding

Transportation is working with two companies that specialize in an end-to-end software development lifecycle, whereby cutting-edge AI tools “would not only build code for engineers, but would actually build an entire application for the organization,” Pidugu said. Specifically, the agency is using Nexgile Nx IDE and NetImpact Strategies’ AI-Forge to “bring the software development life cycle from 18 months to 18 weeks,” or perhaps even shorter spans.

AI’s potential to reduce code developing time is no secret, especially within organizations that have few software coders. At Transportation, Pidugu said the use of AI tools “solved the building-fast-enough problem,” but doesn’t address the full end-to-end lifecycle.

“But we created a brand new problem. We can’t test it fast enough or deploy it fast enough,” Pidugu said, recalling a conversation he had with Jack Albright, now Transportation’s acting CIO. The aforementioned companies were able to leverage AI to demo a product within hours of receiving a business problem from Transportation. The companies further impressed the agency in responding to proof of concepts in a matter of weeks with entire applications and full documentation.

“In general, any vendor that I pick on is usually like, eight to 18 months [to deliver], and a lot of times the documentation never shows up, it’s always incomplete, so these tools being created are really taking it to the next level,” Pidugu said. “I don't have engineers laying around in DOT to take the existing AI models that generate code and use it. So that's why we were looking beyond the traditional vibe-coding methodologies, and these are really good.”

“So what used to take 18 months to write the entire application now is being done in 18 weeks. We could actually bring it down to 18 days,” he added.

What’s next?

Pidugu, who won a Federal 100 award for his work in 2025, has no shortage of opportunities outside government, but said he would like to stay in close connection to public service. He expressed gratitude to Secretary Duffy and Deputy Secretary Steven Bradbury, who — in an interview with Nextgov/FCW, called Pidugu “an amazing leader in terms of IT improvement, development and creating greater efficiencies across the government.”

“It definitely is a great honor to serve the country,” Pidugu said.

In his farewell email to staff, Pidugu fittingly thanked his team.

“Serving alongside this incredible team for the past many years has been one of the most rewarding experiences of my career,” he said. “When we set out on this journey together, we shared a bold vision to fundamentally transform how technology serves the Department and the American public. Thanks to your hard work, dedication, and true builder mentality, we didn't just meet those expectations — we exceeded them. Thank you for your unwavering support, your brilliant ideas, and your partnership. It has been a profound privilege to lead this team, and I will always be cheering for your continued success.”

]]>

CMS hopes to quadruple provider suspensions over next six months, official says

Christian Robles — Thu, 15 Oct 2026 18:38:00 -0400

SANTA FE, N.M. — The Centers for Medicare and Medicaid Services plans to expand provider suspensions and publish the results of a pilot testing whether artificial intelligence can prevent Medicaid fraud, a senior agency official said Thursday.

The two moves are the latest in the Trump administration’s technology-enabled drive to trim fraud, waste and abuse in federal health care programs. The push has helped yield $41.9 billion in savings for fiscal year 2025, or a 59% increase over the previous year, according to a slideshow CMS Deputy Administrator Kimberly Brandt presented at the ADAPT conference in Santa Fe, New Mexico.

In April 2026, CMS launched a “Medicaid War Room” to curtail fraud, waste and abuse. The CMS arm has stopped $342 million in improper payments as of Sept. 2026 and sent 124 notices of intent to exclude providers, according to the slideshow.

“My hope is that suspension number will quadruple over the course of the next six months,” following new exclusion authority granted by the Health and Human Services Department earlier this week, Brandt said.

With that power, CMS can kick providers out of Medicare, Medicaid, TRICARE and other federal health care programs. Brandt previously said she urged congressional offices to give CMS the exclusion authority.

Beyond CMS’ push to exclude certain providers, Brandt noted that the agency is piloting AI and machine learning to detect Medicaid fraud.

The pilots use AI to flag suspicious billing patterns — such as upcoding, phantom services and geographic anomalies — in near real-time before Medicaid payments are made, according to the slideshow. The pilots are also testing a shared platform where CMS, Medicaid Fraud Control Units, program integrity contractors, states and law enforcement work on high-risk cases together.

“If proven out, the goal is to scale this into a national, standardized, continuously improving platform that stops fraudulent Medicaid payments before they’re made,” the slideshow said.

Brandt said CMS will soon publish the results of an Indiana Medicaid fraud pilot the agency conducted with Oracle, and that it is also actively working with five other states on Medicaid pilots. But the agency is also using other technologies to further its mission to combat fraud, waste and abuse.

CMS is utilizing a “Netflix-type algorithm” that compares new providers and suppliers with ones the agency already has a relationship with, Brandt said. The technology allows CMS to put suspicious providers and suppliers on a watchlist for the first six months after they enter a federal health care program. The agency has cut about 67% of those providers and suppliers because they did not follow or comply with program rules, she said.

Earlier this week, the Trump administration announced it cut subsidies for 189,000 people on the Affordable Care Act exchanges after it found those same enrollees had duplicate health care coverage or were dead. The administration said it found some of the duplicate beneficiaries through periodic data matching of Medicaid and the Children’s Health Insurance Program.

Brandt added that the agency is “doing even more” analysis of the Social Security Administration’s Death Master File so as not to issue payments to dead people, though she did not elaborate on what the analysis was.

Even so, CMS’ use of technology to claw back fraud has not been perfect, Brandt acknowledged.

“There are a lot of things about WISeR that are not working,” Brandt said, referring to a controversial pilot of AI-supported prior authorizations of Medicare items and services that have a high risk for fraud.

Hospitals and Democratic lawmakers have criticized the six-year initiative for delaying and denying seniors’ access to care since it launched at the beginning of the year. The House Ways and Means Committee advanced a provision to end the WISeR program in June, while the Senate has rejected efforts to stop it.

“Especially because of the congressional scrutiny on [the WISeR program, CMS Innovation Center staff] are committed to trying to get that right and get it better,” Brandt said.

Among the lessons CMS has learned from the WISeR model are that technology requires continuous oversight, human clinical review is essential and provider feedback is an important monitoring signal, according to the slideshow.

]]>

Everybody hated Socrates

Camille Tuutti — Thu, 15 Oct 2026 16:44:00 -0400

When you ask an AI something, you want the answer. Not eventually, but like, yesterday.

Now, picture one that refuses. You ask what the price of oil will be in six months and it comes back with: What an interesting question! Here's an article and a data set. What do you think?

It's called Socrates, and it was built on purpose to be terrible. It scores zero on every benchmark. Every single person who used it hated it. It also made more of them smarter than any other model in the experiment.

Socrates is the work of Vivienne Ming, a theoretical neuroscientist who chairs the Neurotech Collider Lab at UC Berkeley, has founded seven startups and runs a nonprofit data trust called The Human Trust. I met her in Stockholm this summer, where she'd come to talk about what happens to the people who use AI. She had a cold, a book out — and zero patience for Sam Altman.

Altman had told Congress to imagine a world where every kid has an AI tutor. Ming’s reaction was less charitable than mine would've been. There's a golden rule in the field, she said: If a tutor ever gives students the answers, they never learn anything.

"You know what's exactly like a student with a tutor? A young employee with a copilot," she told the audience at the Future Talent Summit on June 16, in a keynote she'd titled after her new book, "Robot-Proof: When Machines Have All the Answers, Build Better People."

So Ming ran the test, written up in a preprint posted this summer. She pulled questions off the prediction markets, where people bet actual money on what happens next. What will the price of oil be in six months? Will there be a new front in the war in Ukraine? She recruited people around Berkeley, students and adults from the wider Bay Area, put them in teams of three, gave them AI and an hour, and asked for 10 predictions.

She also put EEG caps on them, which told a whole other story. About 60% to 70% of them did the obvious thing, Ming said: They copied the question, pasted it into a chatbot, copied the answer and pasted it into the answer box. Ming calls this group the automators, and she has a less clinical description, too: a very expensive copy-paste function that requires health insurance.

Their brain activity is what worries her. She was tracking gamma band activity, a marker of cognitive effort that drops off naturally with age and drops off pathologically in dementia. The difference between working without AI and working with it, she said, was the difference between doing a hard math problem and zoning out in front of the TV.

Ming has seen it already in the real world. Automated navigation, she told the room, is causally related to reduced memory spans and very likely early cognitive decline. Taxi drivers and ambulance drivers have some of the lowest Alzheimer's rates of any profession, because they have to think about where they're going. Bus drivers don't get that benefit, as they run a fixed route. Neither does anyone following a line on a screen.

"No one on the Google Maps team thought they were building an Alzheimer's machine," she said.

The second group is worse and Ming calls them the validators. They already knew what the price of oil would be, and they asked the model to explain why they were right. The model, being a model, was thrilled to help. Their brains stayed on. Their answers got worse than what the AI produced on its own. The automators only slowed the machine down and the validators actively degraded it.

Between them, Ming told the room, that's 95%.

The last 5% she calls cyborgs. In their transcripts, you can’t tell whether a person or the model made each decision. They beat every human working without AI, and their error was lower than each of the four models on its own. Ten predictions in an hour from people who’d never traded a contract matched a prediction market where the bettors had millions on the line.

It had nothing to do with which model they were handed. Benchmark scores predicted how well each AI did on its own and told you almost nothing about how well the pair did. Cyborgs running a tiny open-source model came close to cyborgs running the state of the art.

What separated them was curiosity, intellectual humility and perspective taking. Did you keep looking after the AI handed you an answer? When it told you that you were wrong, did you ignore it, cave to it, or update and keep going?

I asked her whether any of that can be trained, because she thinks the natural rate of cyborgs in the general population is around 1%. My mother says my most common question as a kid was “but why?” and I’ve made a living asking it since, which makes me a terrible judge of whether anyone else can learn to.

Ming says yes, and then explains what yes costs. Teachers trained to praise 8-year-olds for asking questions instead of getting answers right had kids asking more within weeks and scoring higher on curiosity. West African entrepreneurs who took a crash-course MBA were no likelier to still be in business five years later, but those who went through a resilience program, failing at things and coming out the other side, were 60% likelier to still be open.

In Kingston, Jamaica, researchers spent two years teaching parents social-emotional games to play with their at-risk kids about once a week; twenty years later, those kids earn about 40% more and are far less likely to have been incarcerated.

"The real trick here is, they didn't change the kids," Ming said. "They changed their parents, and the parents stuck with their kids through their childhood."

Which brings us back to the model everybody hated. Ming built Socrates out of the worst-performing AI in her experiment and trained it to withhold. Context and questions only, no answer, ever. The share of users who flipped into cyborg mode went from 21% to 30%.

She's now working on a version people will actually tolerate, which means it eventually has to answer.

"Should I just tell you what the price of oil would be in six months or shall I point you to the resource so you can read it? You'll get the answer, I know you will," she said.

Until a model does that, she does it herself. She uses Google Maps constantly, including in cities she knows cold, because it's the only way to hear about traffic or a closed street. Then she tries to beat it. Maps doesn't know she can cut through a particular courtyard in London. It doesn't know the longer route along the water is the better one.

The Human Trust is building a benchmark for that now, one that scores the person and the machine together instead of the machine alone.

"Wouldn't it be amazing to be able to say our model has the lowest all-cause mortality rate of any model," she said in Stockholm. "The users of our model have the highest income, the users of our model have the most friends."

]]>

AI offers Secret Service unique mission enhancements, official says

Edward Graham — Thu, 15 Oct 2026 15:30:00 -0400

Artificial intelligence is providing the U.S. Secret Service with new capabilities to carry out its protective and investigative missions, the agency’s IT and AI lead said on Thursday, with further adoption of the technologies poised to help it respond even better to future risk environments.

Given the Secret Service’s directive to both safeguard top U.S. officials and visiting dignitaries and to conduct criminal investigations, Chief Information Officer and Chief AI Officer Chris Kraft said “we have to look broadly at what's the art of the possible” when it comes to leveraging AI for its operations.

While Secret Service, like other federal agencies, has leveraged AI to streamline mundane back-office operations or other rote tasks — Kraft noted during the SAP Public Sector Summit in Washington, D.C., that “with generative AI, there's huge opportunity to automate things, such as invoice processing or contracts or other capabilities” — its unique mission positions it to utilize some of the more cutting-edge AI capabilities.

The service is trying to navigate an unprecedented threat environment with an understaffed workforce. During an interview with 60 Minutes on Sunday, Secret Service Director Sean Curran said threats against the agency’s full-time protectees are up more than 40% since last year, with around 10,000 threats made on those individuals since January.

“I believe that artificial intelligence can help us as we investigate those threats, monitor those threats [and] bring together correlations between seemingly disparate information,” Kraft said, referencing Curran’s interview. “I also think when you look at facial recognition and biometrics, [you see] the things that you can do — and at scale — with artificial intelligence.”

Kraft said, for instance, that an object detection tool could help the agency quickly identify how many individuals are in one space “at a speed that a human is not able to do that.”

The Secret Service currently lists three AI use cases in its most recent public inventory, with two of those uses — a model to support criminal investigations through facial recognition matches and another that is used to decipher license plates from poor-quality images or videos — currently deployed by the agency. Kraft said the agency’s AI adoption is meant to empower Secret Service’s human agents to better carry out their duties.

In an interview with Nextgov/FCW earlier this year, Kraft also said the Secret Service is working to embed AI specialists across its operations to help drive further adoption of AI tools.

Kraft highlighted that initiative on Thursday and said developing an internal AI team will allow experts to “go and work with different parts of the organization and accelerate, where it makes sense, our adoption of the technology.”

]]>

Transportation turns to AI to trace research grant recipients’ links to foreign adversaries

Edward Graham — Thu, 15 Oct 2026 14:13:00 -0400

The Department of Transportation has been using artificial intelligence to determine whether grant recipients have ties to adversarial nations, an agency official said on Wednesday.

The federal government has taken a variety of steps in recent years to limit hostile foreign governments’ access to American research, including restricting access to individuals or entities connected with countries like Iran or Russia. As the U.S. seeks to maintain its dominance over China in the global tech race, the Trump administration and Congress in particular have also been tightening the rules on collaborating with organizations or scientists who could conceivably share sensitive technologies or information with Beijing.

It’s not easy, however, for officials or research institutions to unravel these threads, particularly when bad actors are surreptitiously working on behalf of nation states. But new capabilities could help to better chart these connections.

Neil Chaudhry, senior advisor for AI at Transportation, said his agency has worked with partners to create “some really robust AI tools” to analyze the recipients of its research grants.

“When we get to a down-select from grants that are some really promising research projects, we sort of do a deep dive on the [principal investigators] in the grant that says … do they have direct links to the [Chinese Communist Party] leadership, or do they have links to Russian oligarchs?” Chaudhry said during a panel at the SAP Public Sector Summit in Washington, D.C.

“What we don't want to do is go on a fishing expedition,” he said, adding that identifying these kinds of relationships is “not based on the color of their skin and their nationality or something,” but on factors such as whether an individual or entity may be receiving funding from questionable sources.

“If it's a company that's funded by a [venture capital firm] in Silicon Valley, that VC is probably funded by another VC in the Middle East or in Malaysia, and that VC is then wholly funded by the Chinese industrial bank,” Chaudhry said. “And then when that startup goes out of business, we don't know what just happened to that IP that the U.S. government spent money developing.”

Transportation did not respond to a request for comment by the time of publication.

Transportation’s latest AI use case inventory, which was published earlier this year, does not include the research grant-focused capability. Although other agencies have been using automation or more in-depth analysis tools to identify potential connections to foreign adversaries, Transportation’s use of AI seems to be relatively unique.

A November 2024 report from the Government Accountability Office found that reviewed agencies had taken a variety of steps to comply with federal requirements mandating that they better identify foreign risks in the Small Business Innovation Research and Small Business Technology Transfer programs. GAO noted that the programs, which were created to allow U.S. small businesses to engage in federally funded R&D, “face risks of foreign actors seeking to illicitly acquire federally funded research and technologies.”

The watchdog noted at the time that some agencies — like the Air Force, National Science Foundation and NASA — were taking some steps “either to automate the processing of due diligence disclosure forms or support the vetting of applicants and identify associated risks.”

]]>

US moves to staff leaders for industry-partnered offensive cyber program

David DiMolfetta — Thu, 15 Oct 2026 11:09:00 -0400

The U.S. has begun the process of selecting managers to oversee a new program enlisting private companies in operations against foreign cybercriminals, according to a former official and two other people with knowledge of the discussions.

The effort marks an early step toward implementing the initiative unveiled in August. But significant questions remain about how to pay participating firms and who would bear responsibility if their operations cause unintended harm, the former official and one of the people said.

All three sources were granted anonymity to share details of sensitive government discussions. It is unclear who is under consideration for the management roles or whether the government intends to fill them with existing federal officials.

The program allows vetted U.S. companies to gather intelligence on and disrupt systems used by foreign criminal hacker groups. Firms would operate under Justice Department or Department of Homeland Security contracts, with written approval for each mission.

The developments mark a key step toward putting that policy into practice. But the conversations described by the sources highlight how much remains to be settled before companies can undertake the work.

The former official said the government is still working out how to fund the effort and compensate companies. Issuing a presidential memorandum does not itself provide money to carry out its instructions, the former official noted.

“Who’s paying for this to be done, and what’s the financial incentive for any of these offensive cyber companies to actually do this?” the former official said.

Officials crafting the framework are exploring whether companies could be paid based on the success of their operations, but how those payments would work remains unresolved, the former official added. For example, stolen cryptocurrency recovered during an operation may need to be returned to victims, limiting what could be used to pay the companies that helped recover it.

One of the people familiar said companies also need clarity about their exposure if they strike the wrong target, exceed an approved mission or damage systems belonging to someone other than the criminals they are pursuing.

Cybercriminals can route their activity through compromised routers and other equipment belonging to innocent businesses, the person said. Disrupting that infrastructure could require coordination with the FBI and authorities in other countries, including places unwilling to cooperate with U.S. operations.

The U.S. for years has conducted cyber operations, gathered intelligence on foreign hackers and used court orders to dismantle criminal networks, often with technical assistance from private companies. Those activities draw on different legal frameworks, including bedrock Title 10 authorities that underpin military operations and spying.

Firms also regularly help the government investigate and disrupt cyber threats.

CrowdStrike recently provided intelligence for an FBI operation targeting “laptop farms” that allowed North Korean operatives to pose as American remote workers. The White House argues that making greater use of industry’s capabilities under the new directive will help the government pursue those groups faster and on a greater scale.

The former official said managers reviewing proposed missions would have to navigate overlapping law enforcement and intelligence interests. In a hypothetical case, a mission intended to disrupt criminals could interfere with another spy agency’s investigation or access to a network.

Identifying eligible targets presents another challenge. The memo excludes groups that are directly tied to a foreign government, but instructs officials to assume otherwise unless clear intelligence establishes that connection.

As Nextgov/FCW previously reported, that distinction can be difficult to make in Russia, where intelligence services have tolerated, protected or recruited criminal hackers without maintaining full control over their activities. The former official said that ambiguity would place considerable pressure on the managers approving missions.

Participating firms could also become targets themselves. The government would need to consider how to protect companies and their employees from retaliation, including physical threats from criminal organizations, one of the people familiar said.

The initiative has drawn significant attention from industry leaders that have zeroed in on military and intelligence agency demands for exquisite hacking tools and similar innovations that could give the U.S. a leg up in cyberspace.

“It is potentially a transformative approach that imposes cost on the adversary, creating a new form of deterrence, but it’s critical that liability, authorities, and international coordination be sorted out,” Snehal Antani, the CEO of Horizon3.ai, told Nextgov/FCW when asked about the program.

The Trump administration has spent much of the last year defining how private companies would fit into its push for more aggressive cyber operations. Its March cybersecurity strategy called for closer collaboration with industry to disrupt enemy hackers.

But officials’ public remarks on the role companies would have in the efforts have considerably shifted. National Cyber Director Sean Cairncross said shortly after the strategy’s release that he was “not talking about private sector, industry or companies engaged in a cyber offensive campaign.”

Some executives, nevertheless, expected agencies to eventually hire companies to carry out offensive operations.

]]>

OPM revamps online pay tables, spearheads AI adoption across federal workforce

Jory Heckman — Wed, 14 Oct 2026 17:25:00 -0400

The federal government’s human resources agency is revamping one of the most heavily trafficked tools on its website: annual pay tables for federal employees.

The Office of Personnel Management is planning to launch a new version of the online tables that agencies and employees use to look up federal pay rates each year.

Adam Starr, the agency's chief information officer, said on Wednesday that the agency’s pay rates website gets about 14 million visits each year, but hasn’t changed much since 2011.

“I always like to remember that 'OPM' really stands for ‘other peoples’ money.’ And in this case, it is your money. People come to OPM pay tables, asking us to show them the money, and for years these have been legacy, clunky sites,” Starr said Wednesday at the SAP Public Sector Summit sponsored by GovExec. “It’s not the experience that a federal employee deserves.”

Starr said the revamped OPM pay tables will launch “later this calendar year,” in the first quarter of fiscal 2027.

“We’re launching it in beta. This will be the default, but people can always fall back to the old version if they don’t like this new version. We’ll get feedback from the agencies, from the employees. We can iterate, relaunch, and get this in people’s hands. Because that’s how everyone’s doing it in industry, and government should be no different,” he said.

Starr said OPM tested the pay table redesign with an accessibility expert who uses a screen-reader. He said that individual “had absolutely no issues using it.”

“For too long, I think civil servants have really suffered with outdated technology that doesn’t keep at the frontier,” Starr said, adding that OPM is looking to give employees the “latest and greatest tools.”

OPM’s pay tables are the latest in a series of government website refreshes happening under the Trump administration. The agency last year overhauled its federal workforce data website, and the White House’s Nation Design Studio is on a mission to transform more than 27,000 federal websites.

These efforts culminated recently in the launch of America.gov, a site that allows users to ask a chatbot questions about everything from qualifying for Medicare to finding a job to renewing a passport.

Starr described the OPM pay-table refresh as part of a broader plan to “make every part of the federal experience tech-first,” and to get federal employees “the best support that you need in every stage of your career.”

“This is kind of just an example of many of the modernizations that we’re trying to do across OPM, from consolidating 119 different HR IT systems to answering every phone call in our retirement services contact center,” Starr said.

OPM is also spearheading some of the Trump administration’s latest efforts to foster greater adoption of AI tools across the federal workforce. President Donald Trump signed an executive order last month rebranding AI across the federal government as “super intelligence.”

Starr said OPM has launched a consulting program called the SuperIntelligence Governance and Adoption Program (SI GAP), focused on helping other agencies adopt AI tools.

“Every SI transformation industry has HR at the table, and OPM is HR for the government,” Starr said. “Really, it comes down to being able to take measured risk of trusting your employees with these tools, giving them the opportunity to experiment and reinvent their work — because just using it as a better chatbot or a better Google search is not going to reinvent the way that we have the enterprise work.”

Starr said that agencies – including OPM – have historically been too risk-averse with AI, and haven’t yet seized the “opportunity to reinvent their work” with these tools.

“I imagine a world where every employee is more empowered to do the impactful work that they need to do — that a lot of the tedium of the day can be automated,” he said.

Starr said OPM recently launched an AI agent to help process phone calls for the retirement services team.

“We showed it to the entire agency. People were aghast at how human it sounded, how helpful it was,” Starr said. “Literally today we released a new feature, and we released a new feature on Friday last week. And every day this is getting better.”

The federal retirement backlog under the second Trump administration has drawn scrutiny from House Democrats. The backlog sat at 55,681 pending cases on March 31, down from a peak of 65,237 the previous month. The backlog fell to 8,814 pending cases in September.

“We've historically had a large backlog of retirement claims, and it's not because the employees don't want to do good, but there's been processes that are highly optimized on paper that just can't improve until they're digital. They're now digital, and so how can we now optimize those and lead to faster retirements? How can we onboard people quicker?”

OPM in a recent memo directed agencies to use AI to speed up the hiring process for new employees employees. OPM Director Scott Kupor has also been tapped to serve as vice chairman of the administration’s interagency Super Intelligence Task Force.

Starr said OPM understands the importance of keeping guardrails around the data it holds. OPM in 2015 was the target of a massive data breach, exposing about 22 million federal records.

“We have a high risk tolerance for delighting our customers and a low-risk tolerance for security breaches,” he said.

OPM, he added, has three basic rules for AI use: employees are responsible for the work they produce with AI, can use only approved tools, and must stay signed in with official accounts.

“It’s on the CIO team to make sure this is secure, and just constantly thinking about what are the risks and how do we mitigate it. There are concerns about access to the internet via these tools, so we put in a cached version, so nothing hits the internet,” Starr said. “[We’re] just constantly looking at what are the concerns, knocking that down, and iterating and improving.”

]]>

OpenAI bugging case highlights AI’s insider threat problem

David DiMolfetta and John Croxton — Wed, 14 Oct 2026 17:02:00 -0400

An OpenAI employee’s alleged attempt to bug an executive conference room shows how much AI companies have to protect beyond their code, policy experts say, as developers contend with the reality that everyday workplace conversations can reveal sensitive details about their technology plans.

The Justice Department announced Tuesday that Abdulrahman Alfozan, a 35-year-old OpenAI engineer and Saudi national, had been arrested Oct. 9 on a charge of using or attempting to use a device to intercept oral communications.

OpenAI personnel discovered the recorder — concealed in toilet paper and plugged into the back of a television — on Sept. 30. The room hosted discussions about proprietary research, product plans and government engagements, according to the Justice Department.

“We tend to picture AI security as chips, weights, and code. This case is a reminder that the stack also includes the rooms where the roadmap gets discussed and the people who walk into them, and anyone who wants insight into frontier AI has every reason to look at all of it,” said Leah Siskind, director of impact and an AI research fellow at the Foundation for Defense of Democracies.

Prosecutors allege Alfozan had no authority to access the room but was seen near it repeatedly. The FBI complaint names the device as an off-the-shelf camera and audio recording chip which sells for $14 online. They also say security personnel found another similar recorder in a bag containing his identification.

His motive and whether any information was recorded or shared remain unclear, though Alfozan notably had ties to Saudi Arabia’s technology and investment circles. The Financial Times reported Wednesday that he hosted a roundtable at a Saudi-backed conference during IT minister Abdullah al-Swaha’s visit to San Francisco, California last month, and was listed as a venture partner at Riyadh-based Raed Ventures. A court filing also shows he booked a one-way ticket to Saudi Arabia scheduled for earlier this week on Monday.

OpenAI said it fired the employee and had no indication he improperly accessed or exfiltrated model weights, source code or customer data. The company also notably said it used its own AI models to help understand the device and identify who planted it.

Jam Kraprayoon, a senior researcher at the Institute for AI Policy and Strategy, said the operation appeared limited compared with the threats that most concern him.

“This appears to have been a fairly crude operation. It’s still unclear what his motivations were or if he had links to any organized group,” he said. “But from what’s been reported, it also didn’t reach the assets I’d be most concerned about, which are the model weights and the core algorithmic insights behind frontier models.”

Model weights refer to the numerical parameters an AI model learns during training that determine how it processes information and responds. A theft of the model weights could have allowed an attacker to run a full copy of a given AI model, like GPT-6 Astra, on their own servers.

“Recording leadership conversations could expose some high-level algorithmic insights, but this would not be the vector I’m most concerned about,” Kraprayoon added. “The bigger issue is what a more sophisticated actor could do.”

A 2024 RAND study identified 38 distinct ways attackers could target model weights and recommended that developers limit who can access them, keep copies on monitored systems and establish insider threat programs. The study examined threats ranging from opportunistic criminals to well-funded state operations.

Anthropic acknowledged the difficulty in a February update on its Responsible Scaling Policy. Citing RAND, it said meeting the highest security standard — defending against top-priority operations by the most capable attackers — was not yet possible and would likely require help from the national security community.

“The companies themselves and independent analysts have acknowledged that AI developers do not yet have the security posture to protect these assets against a determined, well-resourced actor,” Kraprayoon said.

This is not OpenAI’s only recent brush involving stolen trade secrets. The company also revealed a new filing in its legal battle with Apple that alleged a former employee who left for Apple sent ‌more than 100 files about an unreleased OpenAI product to a personal email account and presented the product to senior Apple leadership.

Siskind said companies should focus on what employees can access and what they do with that access, while continuing to recruit talent from abroad.

“A major part of America’s edge in the AI race is our ability to attract the best talent in the world,” she said. “That being said, we also have to assume that information about every stage of AI development is a valuable target for espionage.”

“Those goals aren’t in conflict if security is built around access, with strong controls on who can reach sensitive systems and real oversight of what they do once they’re in, not just around where someone is from,” added Siskind.

]]>

House Democrats seek watchdog review of AI-powered air traffic control systems

Edward Graham — Wed, 14 Oct 2026 16:05:00 -0400

Almost two dozen Democrats on the House Transportation and Infrastructure Committee are calling for a congressional watchdog to audit the Federal Aviation Administration’s creation and use of new tools designed to provide better airspace supervision.

Transportation Secretary Sean Duffy unveiled the White House's air traffic control 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 effort includes some brick-and-mortar upgrades, much of the push is focused on upgrading the legacy technologies underpinning air traffic control systems.

In a Tuesday letter led by Reps. Rick Larsen, D-Wash. — the panel’s ranking member — and André Carson, D-Ind. — the Aviation Subcommittee’s top Democrat — committee lawmakers expressed concerns about the FAA’s adoption of more advanced capabilities as part of the Trump administration’s broader push to overhaul the systems overseeing the nation’s airspace, and they called for the Government Accountability Office to examine the technologies’ development.

As part of the technology overhaul, the FAA announced a 12-year, $875 million contract with Air Space Intelligence in June to provide two new systems: a Strategic Management of Airspace, Routes and Trajectories — or SMART — tool and Flow Management Data and Services — or FMDS — for a modernized Air Traffic Control System Command Center.

The agency has said that FMDS will replace its current traffic flow management system, with the artificial intelligence-powered SMART tool “serving as an enhancement” within the new system. SMART is designed to combine 200 different data streams into one system to provide better flight traffic oversight, including syncing information like flight paths and weather patterns.

The FAA announced on Sept. 21 that it deployed SMART to monitor airspace in the Washington, D.C. region. That rollout, though, was reportedly met with some initial hesitation from airline executives, as well as Rep. Don Beyer, D-Va., whose district includes Reagan National Airport. One of the criticisms was that air traffic controllers did not have any input in the tool’s creation.

“The use of artificial intelligence within SMART has rightfully raised serious questions about its reliability and safety benefits, particularly on its potential to interfere with decision-making by experienced air traffic controllers or to hallucinate safety-critical information,” the committee Democrats wrote. “These concerns are exacerbated by the Administration’s decision to pilot such a recently developed and technologically complex ATC tool in heavily congested airspace, including the National Capitol Region.”

In an interview with Nextgov/FCW following SMART’s rollout for D.C.-area airports, Beyer said growing global concerns about the capabilities of more advanced AI models led to his office receiving worrisome feedback from constituents about the tool’s deployment. He said, however, that a subsequent conversation with FAA Administrator Bryan Bedford helped to alleviate some of his concerns, including the assurance that SMART was designed by air traffic controllers who aren't in the towers right now but are the “30- and 40-year guys who've been doing it all their lives.”

When it came to SMART’s initial deployment for D.C.-region airspace monitoring, Beyer said “I think the feedback was you almost need big, busy airports in order to generate enough information to make AI relevant.”

Despite FAA’s assurances, however, Beyer and the letter’s signatories have both expressed a desire for further testing and oversight to ensure that SMART is safe.

The House Transportation and Infrastructure Democrats’ letter requests that GAO analyze FAA’s oversight and development activities, including the agency’s safety risk management activities when it comes to both SMART and FMDS. They also called for the watchdog to identify “any quantifiable performance metrics” the FAA is using to monitor the two systems, as well as evaluate the agency’s deployment strategy for SMART and FMDS.

Transportation has pushed back on some of the questions about its use of more advanced systems to enhance the nation’s air traffic control management, with Duffy saying during an appearance on Fox News last month that SMART is “a tool that’s used to better manage traffic” and that air traffic controllers “are the ones who will always manage airspace.”

Duffy has also tried to distinguish between the FAA’s adoption of SMART, and other more powerful frontier AI models that have raised safety and security alarms.

During an event hosted by the National Artificial Intelligence Association in Washington, D.C. last month, Duffy said the air traffic control system upgrades are “going to be driven not by the most complicated AI, but AI that will provide information to [controllers], and will, I think, have a huge impact on the flying public.”

Congress initially allocated $12.5 billion toward the Trump administration’s ambitious air traffic control modernization project as part of the One Big, Beautiful Bill that was signed into law in July 2025. Transportation is seeking an additional $30 billion to finish the wide-ranging initiative, which it hopes to largely complete by December 2028.

]]>