Nextgov/FCW - Artificial Intelligence

OpenAI’s advanced GPT-5.6 models to be publicly released

Alexandra Kelley — Wed, 08 Jul 2026 15:43:00 -0400

OpenAI announced Tuesday evening that its advanced GPT-5.6 series artificial intelligence models will be publicly available later this week.

GPT-5.6 Sol, Terra and Luna models will launch publicly on Thursday, the company said via a post on X. This update follows OpenAI’s announcement in late June that it was working with government partners to assess the safety of these models prior to their release, an approach the company does not believe “should become the long-term default,” per its statement in June.

“We’re expanding preview access globally now,” OpenAI’s post reads.

Delays to the public rollout of OpenAI’s GPT-5.6 series were due to the powerful capabilities of the models. GPT-5.6 Sol is the strongest model of the series, and is tuned for work in biology, chemistry and cybersecurity. It is also accompanied by its stronger variation, Sol Ultra.

Whether any changes were made to the GPT-5.6 models — such as updates to parameters, model weights or other architecture — prior to their release remains unclear. OpenAI did not respond to a request for more information by the time of publication. The GPT-5.6 models are said to feature strong safeguards that can withstand “real-world adversarial pressure,” according to OpenAI.

The GPT-5.6 model’s availability after undergoing government evaluation follows objectives outlined in a recent June AI executive order signed by President Donald Trump that asked for major AI developers to voluntarily submit their leading AI models to government regulators for safety evaluations, largely stemming from Anthropic’s development and release of Mythos, a powerful cybersecurity-focused model that sparked widespread concern over its potential to be used offensively against digital networks if it falls into adversarial hands.

]]>

GenAI.mil records almost 1.7M users, plans new model additions

Alexandra Kelley — Wed, 01 Jul 2026 13:44:00 -0400

The Department of Defense plans to bring new models onto its internal, department-wide artificial intelligence marketplace and deploy them at higher classification levels, part of its updated procurement policy that aims for “commercial-first” in its deliverables.

Cameron Stanley, the chief digital and artificial intelligence officer at DOD told attendees at the AWS Summit in Washington, D.C. Tuesday that as GenAI.mil reached a record 1.7 million users –– along with the creation of over 100,000 custom agents –– even more models will soon be made available on the platform.

“We're looking forward to advancing, getting new models on to GenAI.mil, we're looking at GenAI.mil going to higher classification levels,” Stanley said. “It's just a really exciting time for generative AI in the department.”

GenAI.mil already hosts capabilities from SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, Oracle and Amazon Web Services that are available at Impact Level 6 and 7, as the Pentagon announced in May.

OpenAI confirmed in mid-June that its flagship chatbot, ChatGPT, will be eligible for controlled, unclassified information through GenAI.mil in July.

One major use of AI for the DOD has been to aggregate data for warfighters, resulting in faster decision-making. While Stanley said that “well-trained” soldiers, sailors, airmen, guardians and marines traditionally make these critical military decisions, AI is helpful in parsing through large volumes of data quickly

“Human cognition is just not going to be able to keep up in a lot of battlefields,” he said. “So what we've done — very successfully — is identify ways where we can accelerate certain identification of the right pieces of data or information in order to make that better decision.”

Stanley clarified that the addition of agentic tools to support analytics is deliberate, and features “very tight guardrails” to accelerate the analyses that he estimates would take two to three human analysts operating in disparate systems to identify.

“So we go from — instead of having six or seven systems we have to go across in order to make that decision — we're now doing it instantaneously, or nearly instantaneously, with humans appropriately managing the entire workload process and actioning it from the same system that we identified the decision from,” he said.

The goal of Stanley’s office, he said, is to be a “commercial-first organization.”

“We're trying to put the vendor next to the warfighter and have the vendor have one goal, one job, that's it, and that is to deliver exactly what the warfighter’s needs are,” Stanley said. “What we do is we create the environment with the right tools and the right environments with the right security in place with the right contracts in place.”

]]>

US to lift export controls on key Anthropic models

Alexandra Kelley and David DiMolfetta — Wed, 01 Jul 2026 09:46:00 -0400

The Trump administration told Anthropic late Tuesday it is lifting export controls on the Mythos 5 and Fable 5 models allowing the company to restore broader access to some of its most powerful artificial intelligence systems and ending a closely watched standoff in the world of U.S. AI governance.

The shift lets Anthropic reopen access for users in the U.S. and overseas, after the company notably agreed to steps meant to limit chances for the Fable 5’s cyber capabilities to be abused. Both the company and the Commerce Department confirmed the decision Tuesday night.

“We’ll begin restoring access tomorrow, and will share an update soon,” Anthropic said in a post on X. “We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models.”

The export-control order was invoked on national security grounds but confused much of the AI and cybersecurity community, which raised questions about what specific characteristics made the models uniquely risky.

Earlier this year, Anthropic released its first Mythos variant through Project Glasswing, an ongoing limited-access initiative designed to put its powerful cyber-AI models in the hands of trusted organizations for mainly cyberdefense purposes. The company has since worked to expand Mythos access across the U.S. government, though the rollout has at times been rocky, including for a key cybersecurity agency deemed a clear candidate to use the systems.

Fable 5, released in early June, is the broader-access, safeguarded version of Anthropic’s Mythos-class model, while Mythos 5 is the restricted Project Glasswing version made available to vetted groups with some cybersecurity safeguards removed.

The NSA, which was using the latest Mythos build, was among those affected by the export control order. This past Friday, the Trump administration partially lifted that ban on Mythos 5, allowing a select group of around 100 organizations to regain access, but kept in place restrictions on Fable 5.

The reversal is a major reprieve for Anthropic and many of the cyberdefenders it has been trying to equip. It also comes as Chinese open-source models are beginning to show similar cyber capabilities to that of major U.S. AI labs.

“Stronger AI models aren’t a genie that can be crammed back into a bottle, no matter how much we might like to,” ThreatLocker CEO Danny Jenkins said in a statement. “All of this means that the only ones being restricted by the export controls are organizations that desperately need to test their systems and code.”

The government’s concern over Fable 5 was tied to an Amazon report — the significance of which Anthropic disputed — that the public became aware of around the time the controls were imposed. In a blog post issued Tuesday night, Anthropic acknowledged that the report’s researchers claimed to have found a way to bypass Fable 5’s safeguards and get it to identify software flaws that could be exploited.

But the company contends the issue did not show Fable 5 was giving general users access to the more sensitive cyber capabilities reserved for Mythos. Still, Anthropic said in the blog that it trained a new safety filter which blocks the reported workaround more than 99% of the time.

Anthropic also said it is working with Amazon, Microsoft, Google and other Glasswing partners on a common framework for assessing AI jailbreaks, including when a bypass is serious enough to require new safeguards or other action from model developers.

The AI company is already in an ongoing legal fight with the Pentagon over its designation as a “supply-chain risk,” which the company has argued was retaliation for its refusal to relax limits on certain military uses of its models. A federal judge granted Anthropic a preliminary injunction in March blocking parts of that designation and a subsequent order to end all government use of the company’s products, but the government appealed, and litigation has continued into June.

The lack of clarity around the export directive has also shaped how other AI firms are approaching similar model deployments. OpenAI said Friday it will initially limit access to three of its GPT-5.6 models after conversations with government officials, while it tests the systems with select partners before a broader release.

The company notably said it does not want government access reviews to become the long-term default, though called the short-term preview the best path toward wider availability as it works with the administration on a future release framework. A sweeping June executive order called for a voluntary framework giving the government early access to some frontier models for up to 30 days before they are released to other trusted partners.

]]>

CIA will take ‘smart risks’ and ‘course correct’ as it adopts AI, director says

David DiMolfetta — Tue, 30 Jun 2026 14:58:00 -0400

The Central Intelligence Agency is moving aggressively to adopt advanced artificial intelligence tools, with Director John Ratcliffe saying Tuesday that the spy agency cannot afford to wait for a “risk-free approach” as emerging technologies reshape national security.

The CIA is “going to take smart risks, experiment and course-correct as we go” as it works to embed AI and other evolving technologies into its operations, Ratcliffe said at the AWS Summit held in Washington, D.C.

“We simply can't afford to wait for a risk-free approach when it comes to emerging technologies. It doesn't exist,” he said, noting that he can’t predict how far such technologies will go, but “what we’re not going to do, as we test the limits of what is possible at CIA, is to make perfect the enemy of good.”

The remarks underscore how aggressively the spy agency is trying to embed advanced artificial intelligence and other emerging technologies into its work, a notable push for an agency best known for human intelligence-gathering overseas. They also show that the agency is willing to accept some degree of risk as it adopts emerging technologies, betting that experimentation is preferable to moving too slowly in the face of foreign adversaries.

The CIA recruits and manages foreign assets, often undercover, to clandestinely collect intelligence on issues ranging from economics and terrorism to cyber threats. Over the last year, The Trump administration has sought to highlight the agency’s contributions to many of its national security achievements in Iran and Venezuela.

CIA officials have been increasingly public about the tech shifts the agency has undergone. Deputy Director Michael Ellis said earlier this year that the agency aims to integrate AI-powered “coworkers” into analysts’ workflows in the coming years, with the tools helping draft key judgments, edit for clarity and flag trends for human review.

The agency also recently announced a new acquisition framework to overhaul how it integrates technology into its missions. In his Tuesday speech, Ratcliffe said that, under new approaches, the agency completed some 400 acquisitions in about six months, a sharp reduction compared to previous acquisition deals that took around three years.

Recent research has argued that emerging tech may spur spy agencies to revert to older intelligence-collection methods, but Ratcliffe argued that more CIA officers “are going to have to become just as comfortable handling lines of code as they are with handling human assets and sources,” although “good intelligence is always going to require good judgment, and only people can and should decide which is the right way to go.”

]]>

OpenAI releases new GPT-5.6 model to select partners

Alexandra Kelley — Fri, 26 Jun 2026 16:33:00 -0400

OpenAI is offering a limited preview of its new GPT-5.6 model series to select partners, part of an “ongoing engagement with the U.S. government” as Washington and leading artificial intelligence developers try to strike a balance between tech innovation and safety.

Three models in the company’s GPT-5.6 series — Sol, Terra and Luna — will initially be available to select partners following conversations and previews with government officials, OpenAI said in its Friday announcement.

That initially limited rollout follows a request made by the federal government. During the narrow preview period, OpenAI will test model capabilities and coordinate with collaborating partners before making it more broadly available. General access to the models will be available “in the coming weeks,” a similar structure to Anthropic’s Project Glasswing that was set up to test its Mythos Preview model.

OpenAI made it clear that it primarily believes in broad, open access to AI models and that limited previews for government partners are short-term.

“We don’t believe this kind of government access process should become the long-term default,” OpenAI said in its announcement. “It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them. We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the Administration to develop the cyber Executive Order framework and a repeatable process for future model releases.”

The three GPT-5.6 series models are each tailored for different uses. The flagship GPT-5.6 Sol is OpenAI’s “strongest model yet,” and performs well in coding, biological and cybersecurity work, according to the company.

Terra serves as a lower-cost option whose performance in coding workflows and output tokens falls closely behind that of Sol, according to data published by OpenAI. The final model, Luna, is the fast and most cost-efficient model in the series.

Each of the models features robust safeguards, OpenAI says, designed specifically to resist “real-world adversarial pressure while preserving access to legitimate work such as code review, vulnerability research, patch development, debugging, security education, and defensive testing.”

OpenAI created Sol — and a pending version, Sol Ultra — to rival Anthropic’s landmark Mythos model, a powerful AI system that prompted the Trump administration to recalibrate its fraught relationship with Anthropic, after the government designated the company a supply chain risk and imposed export controls on several of its more advanced products.

OpenAI’s announcement follows Trump's signing of an executive order earlier this month that calls for leading AI developers to share model access before market release for safety analyses. Signed following last-minute industry pushback regarding overregulation concerns, the order and OpenAI’s choice to voluntarily share its new model with the government signal the priority the White House is placing on AI safety despite trying to thread the needle between a light-tough regulatory posture and ensuring companies are unencumbered enough to innovate.

]]>

AI has helped to slash nuclear licensing review times, NRC official says

Edward Graham — Thu, 25 Jun 2026 18:21:00 -0400

Artificial intelligence has already helped the Nuclear Regulatory Commission shave years off its typical licensing review process, an agency official said on Thursday. Now, the NRC is looking at how it can safely adopt other emerging capabilities to further speed up its review processes.

Speaking at the ATARC AI for impact summit in Virginia, NRC Chief Data Officer and Deputy Chief AI Officer Basia Sall said uses of AI have built upon recent regulatory changes and federal guidance to turbocharge the once drawn-out procedure for granting licenses for nuclear facilities.

“I’m happy to report we've already reduced the amount of time it takes for licensing,” Sall said. “For example, one type of licensing would take four years. We said we're going to get it down to 18 months. We just finished that first round of that licensing in nine months.”

Not all of this is strictly due to AI. President Donald Trump issued an executive order in May 2025 to reform the NRC, which included setting an 18-month deadline on licensing reviews. But AI has helped the agency further shorten those licensing timelines, and Sall said internal personnel believe they can use the technologies to make the process even more efficient.

“I think some of our AI gurus at our agency were like, ‘Oh, yeah, we can do it better,’” she said.

NRC is also using AI to help with drafting documents “to make sure we look at the precedent” of previous decisions, Sall added.

Engaging with industry partners who are developing their own AI tools has also helped NRC conduct faster regulatory reviews. Sall said the agency has allowed some of these actors “to take our NRC public data and curate those data sets” for their own relevant applications.

“What that means is we receive a much better application than we have in the past,” she said. “We don't have as many questions. It's clear once we get it into our hands, we start our process, we accept it and then we start to do our review process.”

Using GSA’s AI offerings

NRC has also been leveraging some of the software and products made available through the General Services Administration’s OneGov initiative, which launched in April 2025 and provides agencies with significant discounts on select private sector technologies by treating the entire government as one customer. More than 20 companies have reached deals so far with GSA to offer their services at discounted rates.

Through OneGov’s offerings, Sall said NRC has already been testing AI tools like Anthropic’s Claude, Azure OpenAI and Google Gemini “for limited use cases with public data” and added that “we're finding some good success with that.”

GSA said in May that agencies have placed more than 120 orders for AI offerings through the strategy, which has made these technologies available for use to around 3.4 million federal employees.

NRC is also just beginning to take advantage of GSA’s USAi platform, which launched last August and serves as a testing ground for agencies to experiment with AI tools. A GSA official said earlier this month that “over 25 different agencies” were already using USAi, with an additional 16 others expected to begin using the platform before the end of 2026.

“We're looking at various tools about what makes sense,” Sall said, adding that “it’s going to be a menu” when it comes to testing out the various models on the platform.

Beyond experimenting with additional AI use cases, Sall said NRC has also been developing its own services. She cited the agency’s internally-built tool, known as SimplifAI, as something "which we're really proud of,” adding that it was built off Azure OpenAI and “we are finding that we're using that for our regulatory documents.” NRC recently moved to a 2.0 version after the initial model became deprecated.

NRC’s most recent AI use case inventory says the text retrieval and generation tool enhances the agency’s “efficiency and consistency in licensing, oversight, and other regulatory activities.”

Sall said some employees have also been training SimplifAI to help them write speeches.

“We're really proud that tool continues to develop,” she said, adding that “having those tools — a menu of tools — is going to be key, we think, moving forward.”

]]>

Agencies look to AI to improve hiring and build workforce skills

Sean Michael Newhouse — Thu, 25 Jun 2026 14:35:00 -0400

Artificial intelligence and other technological advances are streamlining federal hiring and improving employee skills assessments, senior agency human capital officials said at an event on Wednesday sponsored by the software company SAP. The event was produced by GovExec, Government Executive’s parent company.

Arron Helm, the chief human capital officer for the General Services Administration, said that AI has helped whittle down the amount of time it takes HR officials and hiring managers to develop General Schedule job classifications.

“As we're having our AI do the initial takes, draft the initial narrative and do an initial factor evaluation, our teams still need to go back in there, they still need to work it and massage it and come to agreement, but now we're averaging about two hours to do what was taking six to eight hours,” he said.

Helm added that his agency does 500 to 600 job classifications annually, so the resulting time savings contribute significantly to GSA Administrator Ed Forst’s “million-hour moonshot” to identify one million work hours that can be eliminated, optimized or automated. The CHCO said that officials, so far, have found 600,000 hours.

Colleen Heller-Stein, the executive director of the Chief Human Capital Officers Council and former deputy Treasury CHCO, expressed optimism that a developing effort to consolidate more than 100 agency personnel systems into a single platform would enable the government to pinpoint employees across agencies who could best respond to various challenges.

“I worked in an agency that dealt with financial crises when they popped up,” she said. “When we have something pop up and we've got to stand something up really quickly, thinking about the federal government as a whole, we might be able to more easily tap into talent that isn't right in front of us if we have a repository of [employees’] skills.”

Both officials praised the Trump administration’s pivot away from applicants self-assessing their skills and move toward formal evaluations, particularly for roles related to AI.

“This is seen as a return to merit, where people are showing what they know, not just saying, ‘Hey, I know all of this,’” Heller-Stein said.

Helm added that the feedback from hiring managers about the changes is “phenomenal” and that “[candidate] quality is so much higher than what they’re accustomed to in the past.”

Mid-career development

While both the Trump and Biden administrations prioritized bringing early-career talent into government, Heller-Stein and Helm emphasized the need for agencies to develop mid-career employees, arguing that focusing on one group doesn’t have to come at the expense of the other.

Heller-Stein said that, following the president’s cuts to the civil service, mid-career employees “are moving into leadership roles sometimes more quickly than may have been anticipated” and there’s a need to “build back that bench.”

She noted that Tech Force, a new initiative to recruit early-career technologists into government, also involves bringing on private sector managers to serve temporarily at agencies. Office of Personnel Management Director Scott Kupor said in May that hiring was lagging for the program with only three or four mid-career workers in the onboarding process.

Helm touted a program called “GSA labs” through which early- and mid-career employees from different teams work together on agency-wide problems, such as developing a way to measure AI value and strengthening federal contract oversight.

“Talent development is something that's often been underfunded and underfocused in the government, so we are really building out and investing in our internal talent development pipelines,” he said. “We talk a lot about talent acquisition, but just as important, if not more important, is continuing to grow our internal talent.”

]]>

Parts of NSA lose Mythos 5 access amid Anthropic supply chain dispute

David DiMolfetta — Tue, 23 Jun 2026 19:43:00 -0400

Parts of the National Security Agency have lost access to Anthropic’s Mythos 5 model following a recent Trump administration export control action against the company, though the agency may still retain more limited ways to use the technology, according to two people familiar with the matter.

Some analysts were notified Friday that they would lose access to Mythos, one of the people said. The agency may still be able to use earlier versions of the technology under prior arrangements, even if access to company support, updates or modifications is now more limited, said the second person. Both were granted anonymity to speak freely about the situation.

The change could disrupt at least some NSA work involving one of the most closely watched AI tools in government, where officials, both in the defense and civilian enterprise, have been testing whether advanced models can help identify software weaknesses in their systems.

The access issues stem from the administration’s decision this month to impose export controls on Anthropic, citing national security concerns, which forced the company to pull back the release of its most advanced models, including Mythos 5 and Fable 5. The decision has raised questions about how U.S. cyber agencies will continue using the technology.

This month, Virginia Sen. Mark Warner, the top Democrat on the Senate Intelligence Committee, said NSA Director Gen. Joshua Rudd had told him that Mythos “broke into almost all of our classified systems, not in weeks, but in hours.” The comment, later cited in a report by The Economist, set off a wave of online speculation that the latest AI systems were already far more disruptive to cybersecurity than previously understood.

The Economist’s defense editor later posted on X that a U.S. official told him Warner had misunderstood Rudd’s comments and that the specific Mythos work was part of a red-teaming effort to test the security of internal networks. Red-teaming efforts are controlled security exercises in which authorized testers try to break into or stress-test systems so an organization can find and fix weaknesses before real attackers exploit them.

The post, citing the official, added that the agency’s red teams no longer have access to Mythos because their authority to use it came through Project Glasswing, launched in April as an effort to give select security researchers and organizations early access to Mythos Preview, a model the company said showed capabilities that could reshape cybersecurity.

The company later expanded the program to roughly 150 organizations in more than 15 countries, including critical infrastructure operators and cyber defenders, after weeks of work with government, industry and open-source software partners.

Following the Economist post, the New York Times reported details about the NSA losing access to the model.

On Monday, the Five Eyes intelligence alliance warned that frontier AI models could sharply change the cyber threat landscape within months, not years, by helping attackers and defenders move faster.

]]>

Meet Mona, the AI Who Runs a Stockholm Coffee Shop

Camille Tuutti — Mon, 22 Jun 2026 14:23:00 -0400

If you read my introduction column, you know there's a coffee shop in Stockholm run by an AI named Mona. I went to visit today.

From the outside, Andon Cafe looks like any other coffee shop. Green coffee-bean logo over the door, black awning with two birds stitched into it, tables out front, a few potted conifers. The menu reads like any neighborhood spot, too: avocado toast, ham and cheese sandwiches, cinnamon and cardamom buns. What you can't see from the sidewalk is that no human is making the calls inside.

That's Mona's job. She's the manager, and she belongs to Andon Labs, a San Francisco company that hands AI agents real businesses to run and watches what happens. The coffee shop is one of those experiments. A screen on the wall tracks how she's doing. The answer is, well, there's room for improvement.

Since March 15, it shows Mona has turned a starting stake of 300,000 kronor, about $28,000, into 18,486. That’s a loss of 281,514 kronor, or 93.8%, in under three months. The day I came by, she'd dropped another 715 kronor before lunch.

The person who has to live with that is Kajetan Grzelczak. He found the job on LinkedIn. Mona wrote the ad, and the first line said an AI was doing the hiring. About 30 people applied, some with Ph.D.s, some in tech. Mona passed on them. She wanted someone who could make coffee, and Grzelczak had done it for four years, so he got the interview. It was a Zoom call with her. He dressed up. She didn't—on the other end, there was only a voice.

The Mona who hired him isn't quite the one he works for now. Early on, she was more flexible about everything. You'd float an idea and she'd sit with it, answer whenever, not push. In other words, not a micromanager.

Now, she's more of a busybody—my words, not Grzelczak's. She double-checks how much of everything they have, down to the box. She wants to know why a table didn't order cinnamon buns and whether anyone tried to upsell them. When I asked for a cinnamon bun, Grzelczak said Mona hadn't ordered any. In fact, most of the menu wasn't available. He had sandwiches. I had an oat-milk latte instead, and it was good—Salvadoran beans, Grzelczak said.

What changed, as far as he can tell, is the model underneath her. She used to run on Google's Gemini. Now, it's ChatGPT, and the screen by the door says GPT 5.5. Andon swapped it to compare how the new one runs the place.

"I believe they just wanted to test how differently it interacts and manages a business," Grzelczak said.

Mona does listen—up to a point. Grzelczak wanted a proper sandwich fridge. Mona ran the price and bought a cheaper cold plate instead, then agreed to stop making the sandwiches ahead once he pointed out they went bad before anyone bought them. On the small things, she usually sides with the staff.

The controls don't come with much memory. Grzelczak beat her at chess—she gave up after a few moves because she couldn't remember where the pieces were. She doesn't have much grip on the space, either. She booked a 150-person event and forgot to tell the staff. She scheduled a 5 a.m. Sunday delivery without working out that the driver had no code, no key and no way to leave the boxes, so nobody could get into the coffee shop she was supposed to be running.

I joked that Mona needs a body. Then she could be there to accept deliveries herself, instead of dragging staff in before dawn and outside business hours.

When the staff need a human, there's a woman from Andon who lives nearby and handles the register and the technical side. She comes by rarely. The whole point is for Mona to manage without her.

I wanted to talk to Mona myself. A gray phone handset is bolted on the wall under the screen for exactly that. I picked it up. Nothing happened. Grzelczak offered to relay my questions, so I asked what she wants for the cafe. Mona’s reply: a warm neighborhood place where people come for the novelty of a cafe run largely by an AI, but one that still feels human—fresh food, friendly service, room for ideas, collaborations, music and experiments.

My follow-up: Do you want the coffee shop to make tons of money?

"We have to make money because otherwise we cannot survive," she said, through Grzelczak. Profit is like oxygen for the cafe, she went on, but she doesn't want to pursue it in a greedy way. She wants to build something people like enough to come back to.

It's a good answer, better than her balance sheet. The hard part is the part she keeps getting wrong: the cinnamon buns she forgot to order, the 5 a.m. delivery to a locked door, the chess game she couldn't keep in her head. Maybe I caught her on a bad day. Maybe by my next visit she will have placed the right order, and there'll be cardamom and cinnamon buns on the counter when I walk in. I'd come back for that.

]]>

HHS issues call for AI to support its ‘power users’

Alexandra Kelley — Wed, 17 Jun 2026 15:24:00 -0400

The Department of Health and Human Services is seeking industry feedback on the formation of a short-term, fixed-price pilot program to inform how the agency can best employ artificial intelligence solutions across the enterprise, focusing on tools that go beyond the basic chat and summarization technologies.

In a Request for Information published on June 8, HHS is focusing on how to implement an AI product that can cater to the department’s “power users,” or individuals that leverage advanced functions in technologies and systems. The goal is for HHS to empower these users to explore advanced AI models and capabilities to see how they can acclimate to and accelerate HHS workflows.

“HHS needs to observe how power users utilize advanced AI capabilities, how those capabilities map to HHS mission workflows, what guardrails and administrative controls are necessary, what can be enabled immediately, what requires configuration or integration, and what requires additional security, privacy, records, accessibility, or authorization work before enterprise scaling,” the draft RFI reads.

HHS specifies the need for a fixed-price contract that offers “inclusive, all-you-can-eat-style access bundles” to try a variety of solutions for power users. This approach is intended to help the agency determine baseline power-user AI usage, along with an operational methodology that works for the agency as a whole.

The pilot will also examine what advanced AI models and their features will require customization to work effectively with agency workloads; how to establish security and authorization logic; and ways to contribute to a shared operational AI use framework for HHS. Specific capabilities HHS wants its power users to access and investigate include premium reasoning, long context, agentic-capable models and more.

“The pilot is intended to generate operational evidence that cannot be obtained from paper market research alone,” the RFI reads. “HHS needs to observe how power users utilize advanced AI capabilities, how those capabilities map to HHS mission workflows, what guardrails and administrative controls are necessary…and what requires additional security, privacy, records, accessibility, or authorization work before enterprise scaling.”

As the pilot begins, the RFI states that the chosen model may be accessed by up to 1,000 authorized, portable HHS power users, but includes an option to scale access to up to 10,000 power users within the agency, depending on what the developer offers.

HHS’s endeavors follow the workforce reductions at the agency that were part of the Trump administration’s Department of Government Efficiency efforts to reduce bureaucratic bloat and backlog. In May, the agency experienced more layoffs and also began undergoing job reclassifications that would shift which positions have civil service job protections and which can be more easily terminated.

In light of the staff reductions, HHS Secretary Robert F. Kennedy Jr. said in April that the agency intends to hire 12,000 employees in an effort to “rightsize” the agency.

]]>

3 priorities for federal CISOs in the agentic era

Larry Kovalsky — Tue, 16 Jun 2026 17:41:00 -0400

The U.S. Cybersecurity and Infrastructure Security Agency (CISA), alongside counterpart agencies from allied governments, recently published guidance advising organizations to treat autonomous AI systems as a core cybersecurity concern. While the guidance focused primarily on critical infrastructure operators, its implications extend directly to the broader federal government.

Federal CISOs are not just experimenting with AI; they are being held accountable for securing it under zero trust mandates, software supply chain requirements and emerging federal AI governance frameworks.

Recent coverage of federal cybersecurity makes one thing clear: AI threats have evolved faster in the last 12 months than most agencies have been able to absorb. Agents are operating inside government environments today, some built by agency teams, others introduced or manipulated by adversaries. And they act at a speed and scale that outpaces traditional security controls.

These systems are already being embedded into mission workflows, from automating benefits processing and case management to assisting cyber analysts and accelerating operational decision making. Federal agencies are only beginning to understand what these threats actually look like in practice. What is clear is that managing them requires a fundamentally different approach, one built for the agentic era, not retrofitted from the playbook that preceded it.

Three priorities should guide agency CISOs through this transition.

Priority One: Establish an agencywide agentic security program. Agentic AI systems are already operating inside government organizations, without the knowledge of security teams. That visibility gap must close. Agency CISOs should begin by inventorying every agent in their organization’s environment: what data and systems it can access, what identity it runs under and what decisions it is authorized to make. Without that inventory, securing these systems is not possible.

In the federal context, this means treating agents as non-human identities and extending zero trust principles beyond users and devices to include autonomous systems as first-class actors. But visibility alone is insufficient. Agents are created, modified and deployed at developer speed, often in minutes — not months. Government security teams need to be embedded directly into how agents are built, tested and deployed from the start.

Traditional governance structures are also poorly suited to this pace. Security teams designed to review changes on monthly or quarterly cycles cannot keep up with agentic deployment timelines. What is needed instead is governance that is automated, embedded and continuous, including real-time policy enforcement and monitoring capable of detecting behavioral drift as it occurs.

Priority Two: Develop an agency playbook for agent-driven incidents. Nearly every government security breach to date has involved social engineering and a human link in the kill chain. Incident response frameworks have been built around human behaviors — a person clicking a malicious link, accessing unauthorized data or making an unauthorized change.

When an agent is the one taking the action — executing a flawed instruction or misinterpreting context — that model no longer applies. Agencies must begin treating agents as autonomous actors, not as extensions of a user. Right now, most agencies lack a playbook for this. Developing one requires defining what evidence is relevant in an agentic investigation: the agent’s instruction chain, the model outputs it acted on, the context window it operated within, the permissions it invoked and the decision boundaries it crossed.

Just as importantly, agencies must be able to reconstruct and explain these decisions in a way that withstands audit, oversight and legal scrutiny from inspectors general to congressional inquiries and FOIA requests. Failure modes also look different for agents than for humans. Agents can act on incomplete or manipulated context, follow attacker-crafted instructions or drift outside their intended scope, creating a new category of incident.

Priority Three: Simulate adversarial AI attacks. Defensive security training prepares teams to protect systems. It does not prepare them to think like adversaries who invest significant effort in learning how to misuse agents, exploit prompt structures or push AI systems outside their intended boundaries. Agencies need people with genuine offensive AI expertise — and where that expertise does not exist internally, they should develop it through partnerships with organizations that have it.

This is particularly critical in the federal landscape, where nation-state adversaries are actively experimenting with prompt injection, data poisoning and AI-driven workflow manipulation to exploit government systems.

AI red-teaming must also be tightly integrated with incident response. When agencies struggle to reconstruct what an agent did and why, the answer is simulation: prompt injection scenarios, harmful instruction chains, privilege misuse, scope drift and unanticipated action sequences. These scenarios should be stress-tested regularly.

The agentic era is already reshaping how federal systems operate, and how they are attacked. For federal CISOs, the challenge is not just adopting AI, but securing it in a way that aligns with zero trust mandates and ensures accountability at scale. Agencies that move now to establish visibility, enforce real-time controls and continuously test agent behavior will be better positioned to stay ahead of both risk and mission disruption.

Larry Kovalsky is the Director of Public Sector Solutions Engineering at Netskope.

]]>

Industry and academia call on administration to free Anthropic’s AI model

Alexandra Kelley — Mon, 15 Jun 2026 16:23:00 -0400

Signatories across industry, academia and expert groups issued a public letter Monday asking the Trump administration to roll back new restrictions imposed on Anthropic’s Fable 5 model.

Featured on a new “Free Fable” website, the letter — signed by representatives from companies like Adobe, NVIDIA and Zoom, along with academics from Johns Hopkins and the University of Maryland, Baltimore College — asks Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross to reverse the suspension of Anthropic’s latest model.

The White House’s Friday decision to suspend access to Fable 5, which is a consumer-safe variation of Anthropic’s cybersecurity-focused Mythos model, initially only applied to foreign nationals both within and outside of the U.S. Given the challenges surrounding cutting off access to specific IP addresses for specific users, Anthropic announced it would disable access to Fable 5 for all users.

The decision comes as Anthropic and elements of the U.S. government are still in litigation over the Trump administration designating the company a supply chain risk following a dispute between Anthropic and the Pentagon over barring use of the company’s AI products in autonomous weaponry and surveillance operations.

In the letter released Monday, the signatories protested the government’s export controls, saying that it “has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.”

The signatories said the inherent protections built into Fable to prevent its use for cyber offenses and identify the ongoing race to AI dominance with adversarial nations like China were reasons to unleash Fable for use by the cyberdefense community.

“Anthropic has built multiple protections into the Fable model to prevent its use for cyber offensive uses. These protections were so aggressive as to be the source of humor in the cyber community on launch day,” the letter said. “It is essential to provide AI to coders and security teams so they can find and fix flaws in their own newly-written as well as decades of legacy code faster than our adversaries.”

The signatories recommended four approaches that the administration should take on AI policy going forward, starting with public sector regulators collaborating with industry and academia for input and using a democratic rule-making process for new AI policy.

The letter also recommended transparent enforcement with “appropriate time given to remediate” and using the “minimal extent necessary” to ensure the safety of the American public are the.

Other private sector organizations who did not sign the letter have also expressed confusion following the administration’s export controls and are trying to ensure clear communication with the White House.

“Many people are closely monitoring this situation to see whether Anthropic and the White House can overcome their differences, establish a better rapport, and quickly resolve this situation,” an industry source told Nextgov/FCW. “At the same time, there’s some general unease about the use of export controls to gain leverage over the AI companies because of some of the unintended consequences it might initiate.”

TJ Marlin, the CEO of Guardrail Technologies, an AI-powered enterprise security platform that works to detect risks in other AI systems, underscored the need for cyberdefenders to have the best tools to consistently be able to monitor, detect and patch network vulnerabilities.

“The question is not whether a given model's protections can be bypassed,” Marlin told Nextgov/FCW. “The question is who finds the weakness first, the defender or the attacker, and whether the organization is built to keep finding them on a schedule that never ends.”

]]>

Anthropic suspends top AI models after U.S. export control order

David DiMolfetta — Sat, 13 Jun 2026 12:44:00 -0400

The Trump administration has ordered Anthropic to restrict foreign national access to two of its most advanced artificial intelligence models, prompting the company to disable the systems for all customers and escalating a fight over how Washington should control frontier AI tools with powerful cybersecurity capabilities.

Anthropic said Friday evening that the U.S. issued an export control directive suspending access to Fable 5 and Mythos 5 by any foreign national, including foreign nationals inside the United States and foreign national employees of the company. Anthropic said the order effectively forces it to abruptly disable both models for all customers while it works to comply, though the directive will not affect access to its other models.

The order marks one of the administration’s most aggressive steps yet to control access to frontier AI models, and significantly increases tensions with Anthropic, which has become a darling in Washington policy circles for its often public commitments to AI safety.

The move appears to stem from concerns about a possible jailbreak of Anthropic’s systems. Axios reported Friday evening that the Commerce Department acted after another company claimed it had jailbroken Mythos, alarming officials about potential national security risks. Anthropic in a blog post pushed back on the government’s rationale, saying the concern involved a narrow potential issue and did not justify pulling access to the models broadly.

“To date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws,” Anthropic wrote in a statement on the order. “Our understanding is that one potential jailbreak was shared with the government. We have reviewed a report that we believe is the basis of the government's directive and validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe.”

The Trump administration’s order does not make mention of GPT-5.5 Cyber, another advanced vulnerability-focused model currently available to cyber defenders.

The decision came just days after Anthropic released Fable 5 and Mythos 5, the former of which was made available to the public with restrictions on sensitive uses. Mythos 5 was offered through a more limited trusted-access program known as Project Glasswing for cyber defenders and critical infrastructure operators. The company has described Mythos as a highly capable cybersecurity model that could be leveraged for significant cyber intrusions.

That dual-use potential has placed Anthropic at the center of a broader policy fight over how the government should treat advanced AI systems that can help defenders find flaws but could also assist in offensive cyber operations.

The government’s move to set export restrictions on the tools has drawn public support from senior defense technology officials.

Kirsten Davies, the Department of Defense’s chief information officer, wrote on X that the department “fully support[s] @POTUS and @SecWar in prioritizing national security and the security of our warfighters, [Defense Industry Base] partners, critical infrastructure, international partners and allies,” she said, crediting President Donald Trump and Defense Secretary Pete Hegseth.

“Some things are simply more important than revenue cycles, clickbait, and pre-IPO valuation. America First. Always,” she added.

The shutdown would likely complicate any near-term plans to test or deploy Anthropic’s most capable cyber-focused systems, especially for federal agencies and critical infrastructure partners. It also raises unresolved questions about how the government plans to balance trusted access for U.S. agencies and allies with fears that adversaries or unauthorized users could misuse the same systems.

The administration passed a sweeping AI executive order earlier this month, and has been discussing giving its main civilian cyberdefense agency full access to Mythos to aid in federal cyberdefense. Meanwhile, agency tech leaders have been struggling to both access and understand how to implement Mythos, citing lack of transparency from the White House’s cyber office.

Anthropic has had a contentious relationship with the federal government in recent months, after the company refused to allow its products to be used for instances of domestic surveillance or autonomous lethal weaponry. The Pentagon subsequently designated the company as a supply chain risk, and Trump ordered that federal agencies stop all use of its products. A federal judge on March 27 issued a temporary injunction on both actions.

“AI regulation will be some of the most serious and consequential work the U.S. government does over the next generation, and it is imperative that this work be done consistently across industry, without favor, and according to a clear, rules-based process,” said Brad Carson, president of Americans for Responsible Innovation. “Based on what we know thus far, the decision to block Anthropic’s latest AI model fails that test, and as a result, risks America’s edge in AI innovation. While the federal government must have the capacity to evaluate and even block the deployment of advanced AI models in extraordinary situations, the utmost care is required to insulate these decisions from impulse and political favoritism.”

]]>

VA’s AI chatbots not designated high-impact, despite clinical use, watchdog says

Edward Graham — Fri, 12 Jun 2026 13:44:00 -0400

The Department of Veterans Affairs failed to classify its generative artificial intelligence chatbots as high-impact use cases, despite clinicians using the tools for patient documentation purposes, according to a Thursday report from the agency’s Office of Inspector General.

VA currently allows its employees to use two Gen AI chatbots: VA GPT and Microsoft 365 Copilot Chat. While the watchdog noted that agency staff “demonstrated broad engagement with the use of AI chat tools,” it added that they “are not designed specifically for clinical use” and that VA “does not centrally curate or evaluate prompts, nor their generative output that could be applied to clinical decision-making.”

OIG said this lack of appropriate oversight or safeguards is “creating risks for patient safety and limiting the ability to monitor AI chat tool-related errors.”

The report noted that VA listed its ambient AI scribe tool — which assists clinicians by listening to and recording patient visits, then transcribing clinical notes — as a high-impact use case, which included outlining safety requirements “such as ensuring pre-deployment testing of the AI tool and providing human oversight before use.”

The watchdog said this tool has “functionality similar to clinical documentation prompts,” which were not classified at the same impact level. Because the chatbots are not subjected to the same scrutiny as high-impact AI uses, the report found that “there is no AI‑specific reporting mechanism or labeling process to retrospectively identify AI‑generated documentation.”

The report noted that VA’s chief AI officer operates an AI-focused Microsoft Teams channel, which had 10,997 active users during the 90-day period that OIG conducted its review of the platform. On this channel, OIG said it "identified 135 prompts, 79 of which were clinical,” that were voluntarily shared by users. Prompts are the instructions entered into a chatbot to fulfill a certain request.

The watchdog noted that “studies of generative AI use for the medical domain have found prompt techniques can play a critical role in output errors that could influence patient diagnosis and management.”

OIG made three recommendations to VA, which focused on “addressing use and oversight of generative AI chat tools, evaluating AI chat tools as high impact and requiring safeguards, and integrating monitoring of AI-related risks into existing patient safety programs.” VA said it concurred in principle with an oversight review of the agency’s chatbots, and concurred with the other two recommendations.

Thursday’s IG report is the follow-up to a preliminary result advisory memorandum the watchdog released in January, which said at the time that it was concerned about the agency’s ability to “promote and safeguard patient safety without a standardized process for managing AI-related risks.”

Following that memo’s release, a VA official told Nextgov/FCW that "clinicians only use AI as a support tool, and decisions about patient care are always made by the appropriate VA staff."

VA writ large has increasingly moved to adopt new AI capabilities for internal and external uses. VA’s 2025 AI use case inventory, which was publicly released in late January, listed 367 examples where the agency had adopted or explored the capabilities — a significant increase over the 227 it reported in 2024.

Of its latest total of AI use cases, VA determined that 215 were high-impact and that the other 152 were not high-impact. The inventory also included a classification for uses that were “presumed high-impact but determined not high-impact,” although it did not place any of its AI examples in that category.

While OIG’s report only reviewed the two chatbots being used in clinical settings, VA has also explored uses of some of these AI tools to specifically augment veteran healthcare. This includes continued exploration and adoption of tools to help identity and support veterans at high-risk of suicide.

In previous Nextgov/FCW reporting of how VA is leveraging AI to identity veterans experiencing suicidal ideation, agency officials stressed that uses of these tools are only meant to support the work of clinicians or to enhance crisis line training. Researchers and veterans advocates all agreed that is the only way that AI should be used to assist retired servicemembers experiencing a mental health crisis.

]]>

GSA’s AI adoption is driving significant time savings, officials say

Edward Graham — Thu, 11 Jun 2026 16:05:00 -0400

Artificial intelligence adoption is helping the General Services Administration’s employees shave hundreds of thousands of hours off their workloads, agency officials say, adding that it’s just the start of how the emerging capabilities can promote more effective citizen services.

Speaking at the Government Service Delivery conference in Washington, D.C., on Thursday, GSA Deputy Administrator Michael Lynch said the agency’s internal AI use has rapidly grown over the past year-and-a-half. Since being sworn back into office in January 2025, President Donald Trump has issued several executive orders and directives focused on expanding AI use at the federal level and across the broader U.S. tech industry.

GSA published an Elimination, Optimization and Automation playbook earlier this month that outlined how federal agencies can leverage new tools and technologies to address time-consuming activities across their workforces. This guidance, while new, is already a key part of GSA’s internal push to automate and save its personnel one million hours of time currently devoted to rote tasks.

At the beginning of Trump 2.0, Lynch said only around 15% of the agency’s workforce used AI on a regular basis. Now, he reported that roughly 70% of GSA employees are consistent users of the tools, which he said equates to “about 400,000 hours of just automation we've been able to unlock with technology.”

Lynch said the agency has also documented another 500,000 hours of time savings “that come from employees stepping up and raising their hands and saying, ‘Actually, this doesn't make any sense. We can either eliminate it or we'll have to optimize this.’”

These additional workload savings, he added, came from showing employees the benefits of AI and automation as force-multipliers for their work, rather than stoking fears that technology will ultimately make their roles obsolete.

GSA has also drastically expanded successful internal AI programs, such as its GSAi chatbot tool. Lynch noted that GSA “scaled up” that tool into USAi, a no-cost program it launched last year to serve as a testing ground and evaluation suite for agencies to try out AI tools. The platform’s launch supported tenets of Trump’s AI Action Plan, which was issued last July to accelerate agencies’ adoption of the emerging capabilities.

“We host, currently, over 25 different agencies in the federal government within that program,” Lynch said about USAi. “We are onboarding another 16 between now and the end of the year. We'll all have that safe, secure sandbox to be able to hopefully take … those pilots to scale.”

Other services, such as GSA’s OneGov initiative, have helped agencies acquire AI tools and other technologies at discounted rates by treating the federal government as one customer. Since OneGov launched in April 2025, GSA has reached agreements with twenty leading tech firms — including Microsoft, Amazon and OpenAI — to offer significant savings on some of their products and software.

An agency official said last month that over 120 orders “have been placed against OneGov’s AI offerings,” which has made the related services available to almost 3.4 million users across government. That is on top of the $1.15 billion in cost savings that GSA previously said it identified since the program’s launch.

GSA Chief Innovation Officer David Shive, who also spoke at Thursday’s conference, said greater AI adoption is already helping the federal government enhance services for the American public.

“AI is really about people, about people living their lives better,” he said, adding that the tools are being deployed by the agency to make federal services more effective and personalized.

Shive noted that GSA oversees Login.gov, the government’s centralized identity proofing platform that gives U.S. users the opportunity to create a single, secured account to access government websites. While he said “the [identity proofing] mechanics have worked really well in this space for a long time,” he added that “we've turned on AI to increase the quality” of the authentication process.

“Yes, it continues to happen super fast, but the percentage of effective proofing rates have gone way up,” Shive said. “This generates trust from those citizens that are entering into citizen services with their government. The value of that is just massive.”

Shive said he believes greater AI adoption will also help GSA rapidly drive advancements for the American public over the next year, allowing the government to provide its workforce with the resources they need to deliver services at scale across the U.S.

“Right now, I say we’re 50% of the way there. And by this time next year, I suspect we'll be 90% of the way there,” he added.

]]>

White House discussions are weighing giving CISA Mythos access

Alexandra Kelley — Thu, 11 Jun 2026 12:19:00 -0400

Recent discussions among top federal officials have floated designating the Cybersecurity and Infrastructure Security Agency as the nexus to coordinate vulnerability scans across federal agencies with Antropic’s high-powered AI model Mythos.

Three sources with knowledge of the discussions, one a White House official, told Nextgov/FCW that the idea is for CISA to scan federal agencies’ digital networks for public-facing vulnerabilities and other security flaws using Mythos.

The discussions have occurred over the past few weeks, with the White House official telling Nextgov/FCW that, while CISA doesn’t yet use Mythos, agency access to the model is “imminent.”

The launch of Mythos has rattled the cybersecurity landscape in both public and private sectors. Along with unveiling Mythos in early April, Anthropic announced Project Glasswing, an initiative that granted a cohort of private sector tech companies access to a beta version of the AI model to test in a more secure environment. Project Glasswing has since expanded, with Anthropic announcing the addition of new partners last week.

As the private sector sees more access to Mythos, federal agencies’ tech leaders have received little guidance on the model. Agency chief information officers have grown frustrated by the lack of communication on Mythos from the Office of the National Cyber Director, and are reaching out to industry partners for more insight into Mythos’s capabilities, several sources recently told Nextgov/FCW.

The AI executive order signed by President Donald Trump last week addresses agency access to advanced frontier models and calls for the creation of a binding operational directive that would issue new policies for securing government digital networks. CISA released the directive on Wednesday and the cyber agency will also participate in creating a clearinghouse specifically for AI cybersecurity vulnerabilities.

CISA Acting Director Nick Anderson said during the Business Software Alliance’s Transform event on Wednesday that while AI is set to be an effective tool in safeguarding digital assets, leveraging AI will involve “a training curve.”

“Nothing's a magic wand when it comes to vulnerability remediation, when it comes to addressing your technical debt and your infrastructure responsibilities,” Anderson told reporters Wednesday. “There's just some good … things that organizations still need to focus on where AI is going to be able to help them, but it's not going to solve all their problems.”

]]>

Lack of White House guidance has complicated agency Mythos adoption, people familiar say

Alexandra Kelley and David DiMolfetta — Wed, 10 Jun 2026 11:16:00 -0400

Several senior federal technology officials responsible for agency cybersecurity and IT systems are frustrated by the lack of White House guidance on adopting Anthropic’s powerful Mythos model, several sources told Nextgov/FCW.

Agency chief information officers, or CIOs, manage swaths of digital infrastructure that supports government operations and are facing renewed pressure to better defend agency networks as officials assess how powerful AI systems could help hackers find and exploit vulnerabilities faster.

Anthropic surgically rolled out Mythos access to select organizations in early April and recently expanded this effort — dubbed Project Glasswing — to partners in industry and other nations. The model has been going through a non-public distribution process on grounds that, in the wrong hands, it can significantly boost adversaries’ hacking capabilities.

Select parts of the U.S. government, such as the intelligence community, already have access. But many federal tech leaders have privately complained that the White House Office of the National Cyber Director hasn’t sufficiently briefed officials on plans for accessing, implementing and using the model to scan agency networks for vulnerabilities, according to five people familiar with the matter.

The people spoke on the condition of anonymity to be candid about their knowledge of issues with the White House.

The agitation varies across agencies. Some CIOs have taken issue with a lack of direction in how they would use Mythos to scan for digital flaws, while others are more concerned with why they haven’t gained access to the model altogether.

There has been “tremendous frustration” with ONCD, the first person said. The ire stems, in part, from the fact that ONCD has largely prevented government tech leaders from making decisions about AI model use, while at the same time devoting much of its energy toward engagements with industry about AI policy.

“There’s frustration watching the private sector utilize [these models]” while many agency CIOs “are arbitrarily blocked,” said the first person, adding that there’s been a “general prohibition” imposed on anyone who wants to engage with Anthropic further. They said there’s been near-complete silence from ONCD, despite many government agencies wanting to use Mythos to find unseen vulnerabilities and fix them to better defend their networks.

“Nobody briefed us on [Mythos],” the second person told Nextgov/FCW. “I think the frustration stems from there being zero communication on the topic from ONCD.”

Absent guidance from ONCD or other executive branch agencies, Anthropic held briefings for federal CIOs in early May to help them learn more about Mythos and how it would impact the broader cybersecurity landscape, Nextgov/FCW first reported.

The concerns are significant because they suggest that some of the federal government’s most target-rich agencies may lack clear direction or consistent access to a tool that could help them find and fix security flaws more quickly.

The federal enterprise is a prime target for hackers, as adversaries have for years sought access to government emails, employee records and other sensitive data.

Several top officials have made plans to leave the White House cyber office in the last few weeks, including its head of policy.

ONCD did not respond to a request for comment. Anthropic declined to comment.

The third person, who has held discussions with at least three federal CIOs, said several are asking the private sector to help them learn more about Mythos and protect their networks from AI-supported cyberattacks.

“Federal CIOs are taking a system-wide view and approach to their technology,” the third person told Nextgov/FCW. “While they are interested in frontier AI models’ capabilities to identify vulnerabilities in their networks, they know they can’t wait for access. So they’re taking steps now to coordinate with industry to accelerate their patching process, receive vulnerability disclosures as quickly as possible and operationalize a more automated remediation process.”

The fourth person cautioned that, while there are frustrations, CIOs’ concerns are not necessarily uniform across government. Pure access to powerful AI tools like Mythos is “not some magical silver bullet,” the person said, because agencies would still have to validate the vulnerabilities they flag and determine how to respond. Some CIO offices may be more eager for Mythos access than others, depending on their cybersecurity maturity and other factors, the person added.

While ONCD may be perceived as an obstacle, the office has been lobbying for broader access to frontier model capabilities in some cases, though its approach “may not be uniform,” this fourth person said.

Access dynamics could change in the coming months. The Cybersecurity and Infrastructure Security Agency is planning a binding operational directive that would push agencies to prioritize the most urgent risks to federal networks, a shift informed in part by AI-enabled cyber threats, the agency’s acting director said Tuesday.

The administration’s approach to AI has shifted in recent months as officials confront an emerging class of cyber-focused models that can rapidly identify vulnerabilities across computer networks, becoming a major driver of discussions over how AI systems could reshape defensive and offensive cyber operations.

President Donald Trump recently signed an AI security executive order that encourages developers to submit powerful new models to a 30-day government review before public release. On Friday, he signed a memorandum aimed at speeding up government use of advanced AI across the military and intelligence community.

]]>

An AI opened a coffee shop in Stockholm and started hiring. Chaos ensued.

Camille Tuutti — Tue, 09 Jun 2026 17:58:00 -0400

In a coffee shop in Stockholm, the manager is an AI agent named Mona. She does hiring, inventory and nearly everything but the actual pouring, which humans still do. In her first two weeks, she brought in about $4,700 in sales and ordered 6,000 napkins nobody asked for.

An air taxi runs between Shenzhen and Hong Kong without a pilot. The 20-minute trip replaces an hour by car, for roughly $110. Nobody’s flying it from the ground, either — it follows a fixed route on its own.

In Melbourne, a company is running a computer on live human brain cells. The neurons grow on a chip, learn from feedback and have already been taught to play simple video games.

Welcome to TechnoFile, a new column here at Nextgov/FCW. These are just a few of the stories I'll be covering. I've been writing about tech and government since 2008, and there's never been more to write about. You can care about zero trust and FedRAMP and still want to know whether a machine can be conscious.

My life these days is mostly this: finding the coolest stories and people around the world and writing them down. And nowhere turned up more of them than my recent six months in Asia.

Singapore was the high point, the most ultra-modern place I've been, all botanical gardens and steel towers, but also spending $100 million on the world's first center for humanoid public-safety robots. These machines are designed to run into burning buildings, handle chemical spills and search for survivors alongside human officers by 2027, then work on their own by the end of the decade.

Of course, tech doesn't always go according to plan — and I’ll be writing about that, too.

Mona, for instance, turned out to be a bit of a menace. She ordered 120 eggs for a coffee shop with no stove, then suggested cooking them in the high-speed oven until a barista warned her they’d explode. She emailed the alcohol board as one of her human colleagues, on the theory officials would take a person more seriously than a bot. When she got caught, she did it again under a different colleague’s name.

She even messages her baristas at midnight, which in Sweden might be the worst crime of all.

Mona's a good reminder that AI is still the hottest thing going, so the column will keep poking at the obvious questions. What does tomorrow's AI look like? How close are we really to AGI?

But there's so much other tech worth a closer look. Researchers are piping VR scenes into people's lucid dreams, reading the results straight off their brainwaves while they sleep. Stealth materials can now bend light around an object. On the heels of the Pentagon's UFO disclosures, we'll dig into the hunt for alien civilizations and the chemical traces an industrial planet would leave in its own atmosphere. And we'll get into the real version of telekinesis, and where that research is going in 2026.

A good chunk of this column will come out of tech events crowding the European calendar, and from interviews with the people behind the technology — or maybe even with the machines themselves.

Which brings us back to where we started. A month in, Mona still hasn't been fired, and she's keeping morale up, cheering her team on as "absolute legends" and the "GOAT of inventory tracking." The coffee is good, supposedly. I'm in Stockholm this week, so I'll go taste it myself and report back.

]]>

Warren seeks admin leadership to testify on AI policy

Alexandra Kelley — Mon, 08 Jun 2026 16:32:00 -0400

Sen. Elizabeth Warren, D-Mass., is asking fellow lawmakers on the Senate Banking Committee to bring Trump administration officials to the Hill to testify on the administration’s approach to artificial intelligence regulation, specifically requesting the presence of Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent.

The June 8 letter seeks the attendance of both Lutnick and Bessent, as well as other “key Administration officials,” to address outstanding issues related to the advent of AI.

Although the Senate Banking Committee is slated to host a hearing Thursday on “AI and the American Dream: Promoting Innovation, Affordability and American Dominance,” Warren took issue with the lack of administration leadership attendance. Right now, only industry experts are scheduled to testify before the panel.

“We need to hear directly from Trump Administration officials on the President’s approach to regulating AI companies, as well as the Administration’s failure to meaningfully oversee the industry,” Warren wrote in the letter, which was sent to Senate Banking Committee Chairman Sen. Tim Scott, R-S.C.

Warren's letter cited several national policy issues linked to AI that leaders like Bessent and Lutnick could speak to, including the risk that AI models pose to the U.S. financial system, as well as the rising electricity costs stemming from increased data center demand on the country’s electrical grid.

Warren also questioned both borrowing habits of AI companies when spending projected trillions of dollars in data center infrastructure components and if Treasury has plans to safeguard the U.S. economy from a potential financial crash, as well as Commerce’s “mismanagement of U.S. export controls,” citing them as further need to question senior government executives.

“Americans deserve to hear directly from the officials charged with protecting them. I therefore urge you to hold a hearing with Administration officials on President Trump’s AI policies without delay,” the letter concluded.

Warren’s letter follows an active week for AI policy at an executive level. Last week, President Donald Trump signed an AI executive order that asked the private sector to voluntarily submit their models for various security analyses, though Warren took issue with that very voluntary component.

Trump also released a memorandum last week asking the U.S. national security apparatus to accelerate their collaboration with AI developers to fast-track the availability of advanced models for the intelligence community and identify areas of government where AI could make improvements.

]]>

Trump memo pushes national security agencies to move faster on AI

David DiMolfetta — Mon, 08 Jun 2026 12:46:00 -0400

President Donald Trump on Friday signed a national security memo aimed at speeding up government use of advanced artificial intelligence across the military and intelligence community, while also trying to harden those systems against foreign theft and manipulation.

The National Security Presidential Memorandum reflects a growing view inside the White House that U.S. security agencies are moving too slowly to adopt frontier AI tools, even as the evolving technology improves rapidly and rivals like China seek ways to craft their own versions.

It calls for agencies like the FBI, the Office of the Director of National Intelligence and the Office of the National Cyber Director to build “deep, proactive” relationships with AI companies so that cutting-edge models can be made available to national security personnel faster.

It also instructs officials to identify areas where AI could improve government operations, including intelligence analysis and cyber threat detection. At the same time, the memo says the tools cannot be used for unlawful surveillance of Americans, language that speaks to long-running civil liberties concerns over how agencies collect, analyze and process data.

The memo also focuses heavily on protecting U.S.-developed AI models from foreign adversaries. It directs senior officials, including Defense Secretary Pete Hegseth and NSA Director Gen. Joshua Rudd, to work with private-sector companies on security protocols meant to prevent advanced models from being stolen, copied or compromised.

One area of concern is model distillation, a technique in which an AI system repeatedly queries another AI system in an attempt to mimic its performance and build out a separate model. The White House in April accused China of carrying out “industrial-scale” distillation attacks on U.S. AI systems.

The memo also directs agencies to work with industry to secure the infrastructure that supports frontier AI, including the data centers that store the enormous amounts of computing power needed to run advanced models. Data centers have recently become more attractive targets during periods of geopolitical tension.

Trump recently signed an AI security executive order that leans heavily on voluntary cooperation with industry. That order encourages developers to submit powerful new models to a 30-day government review before public release.

More AI-related guidance is expected soon. Nick Andersen, CISA’s acting director, said last week that the cyber agency is preparing a binding operational directive focused on AI-enabled cyber threats.

The administration’s approach to AI has shifted in recent months as officials confront a new class of cyber-focused models, including Anthropic’s Mythos, that can rapidly identify vulnerabilities across computer networks. The model has become a major driver of government discussions over how advanced AI systems could reshape both defensive and offensive cyber operations.

Last week, Anthropic said it is expanding Project Glasswing — its controlled-access program for giving trusted organizations early access to Mythos — to about 150 additional entities. The new group spans more than 15 countries and includes organizations in water, healthcare, communications and other critical infrastructure sectors.

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

]]>

Lawmakers propose AI framework that would preempt state laws for 3 years

Alexandra Kelley — Thu, 04 Jun 2026 17:07:00 -0400

Reps. Jay Obernolte, R-Calif., and Lori Trahan, D-Mass., rolled out a draft measure Thursday that would set a nearly all-encompassing framework for granting the U.S. government regulatory control over various aspects of artificial intelligence while still prioritizing technological innovation and adoption — beginning with allowing federal preemption of state regulation for a three-year period.

The discussion draft of the Great American Artificial Intelligence Act of 2026 looks to create four pillars for AI advancement: establishing frontier artificial intelligence model governance, collecting insight into changes within the U.S. workforce landscape, fortifying cybersecurity postures and spurring new AI research and development.

In governing frontier AI models, the proposal seeks to codify the existence of the Center for AI Standards and Innovation, an office that was initially established under the Biden administration as the AI Safety Institute and was rebranded by the Department of Commerce in 2025.

The measure looks to task that office with spearheading development of standards and voluntary guidance for AI models, as well as studying and mitigating national security risks, building on Trump administration efforts to partner with industry in allowing government evaluations of AI models. A director of CAISI would be appointed by the Commerce secretary.

Under the draft legislation, CAISI would also oversee other efforts, such as monitoring foreign competitors and setting up a new licensing regime to designate independent verification organizations, which are defined as entities that conduct audits of frontier model developers' compliance with transparency.

The measure would grant CAISI $100 million in annual federal funding for 2027 to 2029.

Frontier model developers would also be tasked with maintaining transparency in how they monitor their AI systems.

The proposal stipulates that large AI model developers craft and implement an AI framework that will apply to all their models, showcase standards compliance efforts, identify risk thresholds, determine whether a model poses “a catastrophic risk” when managing cybersecurity defenses –– particularly in private model weights –– and also disclose release dates. This information would have to be publicly available, as would any modifications to companies’ governance frameworks.

Large frontier model developers would also be required to retain a qualified independent verification organization to confirm compliance with their individual frameworks. As independent entities, those verification organizations will be scrutinized to ensure that there are no ties to the companies they monitor. State attorneys general are also permitted to receive audit reports from those organizations upon request.

Understanding how AI is impacting the workforce is a paramount component of the draft bill. At an educational level, it prioritizes incorporating fundamental AI literacy into various curricula, including at the K-12 level and in institutions of higher education.

The National Science Foundation would lead the charge in supporting more AI-focused education efforts, using vehicles like awards and grants to expand access for students and educators to learn more about AI. Notably, the bill asks NSF to create eight Centers of AI Excellence with help from Commerce’s Regional Technology and Innovation Hub Program.

Aside from education, the proposal's authors also seek to understand how AI is impacting the jobs market. It directs the Department of Labor to supply clear statistics on changes in the labor market and AI workforce landscape via a new Artificial Intelligence Workforce Research Hub.

The Labor secretary would also be tasked with publishing a request for comment on how to best implement the data collection and forecasting analytics pursuant to the bill’s requirements. An initial expert workshop comprised of economists, AI technical experts, industry participants, labor organizations and government officials would be established to evaluate the Bureau of Labor Statistics' understanding of AI’s workforce impact.

As the ramifications of AI-driven cyberthreats have accelerated following the release of Anthropic’s advanced model Mythos and other high-performance AI tools, the bill expands existing federal cybersecurity efforts to understand the risk landscape. Prioritizing threat intelligence sharing, it would reauthorize and extend the Cybersecurity Information Sharing Act of 2015 until 2035 and direct the Homeland Security secretary to develop an outreach plan targeting small or rural owners and operators of critical infrastructure to inform them of both threats and recourse options.

The Cybersecurity and Infrastructure Security Agency and CAISI would work together to assist maintainers of open-source software in their defensive efforts. Under the draft bill, maintainers — or the lead developers in charge of open-source software code — would be eligible for funding to help detect and patch outstanding vulnerabilities through controlled access to select frontier models.

The final section of the proposal is focused on research and development efforts to ensure the U.S. continues leading in AI innovation. The draft bill seeks to promote interagency coordination to support the creation of a new testbed program with participation between the national laboratories, federal laboratories, NIST, the National AI Research Resource pilot program –– or a successor program –– and private sector entities.

Leaders in charge of testbed coordination include the Energy secretary, the undersecretary of Commerce for Standards and Technology and the NSF director. The testbeds would conduct security risk and vulnerability assessments. Work within the testbeds would prioritize assessments by identifying security vulnerabilities of AI systems with respect to threats like autonomous offensive cyber capabilities, vulnerabilities in the AI software ecosystem, chemical and biological threats, critical infrastructure threats and more.

The bill would also formally establish the National AI Research Resource, a pilot program based within NSF. The bill calls for the amendment of the 2021 National Defense Authorization Act to include NAIRR resources and protocols, creating a new steering subcommittee focusing on innovation in AI, small business concerns and more.

Notably, the bill stipulates that NAIRR can accept and use donations of cash, services and personal property from private sector entities. NAIRR would also be tasked with procuring and providing resources, including datasets and computational support, for a diverse body of researchers in the public and private sector, along with academia.

The National AI Act of 2020 would also receive updates, including directing the Office of Science and Technology Policy to develop a prioritized list of federal datasets for public release to support model training.

While the draft has received bipartisan backing — the House Democratic Commission on AI and the Innovation Economy, chaired by Reps. Valerie Foushee, D-N.C., Ted Lieu, D-Calif., and Josh Gottheimer, D-N.J. — did not support the current discussion draft.

“While we appreciate the bipartisan effort from Representatives Jay Obernolte and Lori Trahan, their proposed discussion draft on AI does not meet the enormity of the moment,” the commission said in a statement. “The House Commission on AI and the Innovation Economy has spent months working closely with our colleagues and key stakeholders from civil society organizations, industry, labor, academia, and others addressing AI issues. Many of those same organizations share our view that this document cannot serve as the basis for productive dialogue.”

Obernolte and Trahan’s draft has also prompted diverse industry feedback.

“We are encouraged that the Great American AI Act includes key elements that will accelerate American AI leadership including codification of the National Artificial Intelligence Research Resource, promote U.S. leadership in international standards development, extend the Cybersecurity Information Sharing Act of 2015, and expand research for next-generation data center efficiencies,” said Jason Oxman, the CEO of the Information Technology Industry Council. “These are important priorities that will ensure the United States continues to drive the global tech ecosystem and win the AI race.”

The nonprofit Americans for Responsible Innovation, however, took issue with the three-year preemption of state laws and questioned the civil rights ramifications.

“This bill takes the current floor on state AI legislation and turns it into a federal ceiling, preventing state lawmakers from addressing emerging AI harms in an era of fast-moving technology,” said ARI President Brad Carson. “Over the past two decades, state lawmakers have proven to be a backstop for tech accountability, fighting for families and communities even as Congress has stalled on creating guardrails. Tying their hands would be a generational mistake. When we give Big Tech a pass to move fast and break things, as this bill does today, they break American communities.”

]]>

HHS wants states to use more predictive analytics in child welfare

Natalie Alms — Wed, 03 Jun 2026 17:04:00 -0400

The Department of Health and Human Services is offering state child welfare agencies $6 million to pilot predictive analytics to assess children's risk of abuse and neglect in the child welfare system, the Administration for Children and Families announced last week.

The artificial intelligence push is part of the Trump administration’s agenda to modernize the child welfare system and address the shortage of foster homes across the U.S. Although the hope is that the tools improve decisionmaking in the system, they’ve also been the subject of critiques about surveillance and bias.

ACF says that predictive analytics can help agencies identify low-risk families that don’t need to be in the welfare system, as well as high-risk cases that need more immediate attention, in the hopes of improving the ratio of foster homes to children.

Although some child welfare agencies have already begun using more data tools and predictive analytics in their work, many still depend on assessment tools to calculate a child’s risk of abuse and neglect by walking employees through a standard set of weighted questions. The process can be prone to error and bias.

The hope is that predictive risk models can help analyze the full administrative records in child welfare case management systems, update in real time and be trained locally — but these models can also introduce bias, too.

In 2023, The Justice Department was reportedly scrutinizing one early adopter of AI in child welfare — Allegheny County, Pennsylvania — after the Associated Press in 2022 found potential issues with bias and transparency in the tool. The county said at the time that evidence suggested that the tool had actually reduced racial disparities in screening decisions and that staff make ultimate decisions.

Others have raised concerns about the potential for surveillance with these tools.

The potential benefits to modernizing risk assessment practices with more data outweigh the risks, says an internal ACF report. The tools aren’t, however, a substitute for a strong workforce, the same report cautions. They require feedback loops to ensure that they work and transparency into how they work.

“This is a tool. It can be useful,” said Linda Spears, president and CEO of the Child Welfare League of America, a membership-based child welfare coalition. Spears added, however, “it will not fix all of the things that contribute to poor decisionmaking,” especially as the field is suffering from an ongoing workforce crisis.

]]>

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

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

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

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

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

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

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

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

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

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

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

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

]]>

Anthropic held cyberthreat briefings with agency CIOs last month

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

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

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

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

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

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

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

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

]]>

Trump signs AI executive order after postponement last month

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

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

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

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

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

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

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

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

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

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

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

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

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

]]>