White House drafts AI policy memo as Anthropic dispute intensifies over Mythos expansion

White House AI Memo Hits Issues Driving Anthropic-Pentagon Feud

The memo's language addresses some concerns raised by the Pentagon and Anthropic during contract negotiations, and some officials think it could offer a path for the Pentagon to voice concerns about Anthropic while keeping the firm's technology in military systems. White House officials are preparing a wide-ranging artificial intelligence policy memo that outlines requirements for AI deployment by national security agencies, some of which touch on issues driving the bitter dispute between the Pentagon and Anthropic PBC over military use of the firm's technology, according to people familiar with the matter. In the works for months, the draft memo urges US agencies to use multiple AI providers to avoid the vulnerability of relying on a single vendor, according to the people, who requested anonymity to discuss ongoing deliberations. It also calls for AI companies that contract with the Department of Defense to agree not to interfere with the military's chain of command, where the president has the final say, the people said. The memo, which is subject to change and covers an array of related priorities, is intended to replace the Biden administration's national security memorandum for AI and isn't specific to Anthropic or the Pentagon, the people emphasized. Still, with its call for a range of AI vendors, some officials think the document could offer political cover for the Pentagon to quietly back down on its claim that Anthropic -- until recently the only AI company approved for classified work -- posed a threat to the supply chain by insisting on safeguards for use of its products. A senior White House official, who spoke on condition of anonymity owing to the sensitivity of the matter, said the memo isn't designed to create any kind of off-ramp in the Pentagon's conflict with Anthropic. Rather, the official said, the administration aims to onboard other vendors as quickly as possible, so that "when" Anthropic is removed from government systems, the US has the necessary technology to support military operations. The official added that the administration wants to work with all American companies in the long term. Asked to clarify the White House's position on Anthropic, the official referred to President Donald Trump's social media post calling for federal agencies to cease use of the company's technology. Spokespeople for the White House, the Pentagon and Anthropic declined to comment. Axios first reported that the White House was working on guidance that would allow government agencies to "get around" the Pentagon's designation of Anthropic as a supply chain risk. The memo doesn't lift that declaration. But its language, details of which have not previously been reported, addresses some of the concerns raised by the Pentagon and Anthropic during increasingly tense contract negotiations that ultimately broke down over the issue of guardrails. In a nod to the Pentagon's priorities, the people said, the memo would require that AI models used by the military aren't changed without government permission and are free from ideological bias. It also affirms that AI companies must strictly adhere to the chain of command -- but stops short of requiring that companies agree to "all lawful use" of their products, which is the specific language the Pentagon has demanded in military agreements. Other sections of the document touch on key areas of concern for Anthropic. Those include a requirement for defense and intelligence agencies to add language in contracts reinforcing constitutional protections and assuring vendors that their AI technology wouldn't be used for unauthorized surveillance, according to people familiar with the memo. The document would also mandate annual updates of Pentagon regulations governing autonomous weaponry, the people added. Taken together, some officials think those terms offer a path for the Pentagon to voice concerns about Anthropic while keeping the firm's technology in military systems alongside providers like xAI and Alphabet Inc.'s Google, the people said. Defense Secretary Pete Hegseth had outlined a six-month transition period for the Pentagon to phase out Anthropic, whose Claude Gov AI tools are used by Palantir Technologies Inc. in the Maven Smart System, an AI-enabled mission control platform that US armed forces have used in operations against Iran. It could still be a messy road ahead as multiple lawsuits by the company play out. Earlier this month, a federal appeals court declined Anthropic's request to pause the Pentagon's supply chain risk designation, even as plans for a broader government ban on its technology remain temporarily blocked by a California judge. White House officials had been working behind the scenes to untangle the feud ahead of the Pentagon's late February deadline for Anthropic to drop its demands for guardrails, people familiar with the matter said. Vice President JD Vance, Office of Science and Technology Policy Director Michael Kratsios and National Cyber Director Sean Cairncross were among those trying to cool tensions, the people said. Even Trump's statementBloomberg Terminal on Truth Social, posted shortly before the Pentagon's ultimatum expired, made no mention of supply-chain risk, though he did order government agencies to sever ties with Anthropic. Get the Tech Newsletter bundle. Get the Tech Newsletter bundle. Get the Tech Newsletter bundle. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Plus Signed UpPlus Sign UpPlus Sign Up By continuing, I agree to the Privacy Policy and Terms of Service. The search for a solution accelerated this month following Anthropic's announcement that its new Mythos AI model could pose a global cybersecurity risk, according to people familiar with the matter. On April 17, White House Chief of Staff Susie Wiles and other administration officials met with Anthropic Chief Executive Officer Dario Amodei at the White House, where topics discussed included Mythos, the people said. Wiles has become more involved in AI policy, including with the memo, the people said. The White House memo is taking shape as the Trump administration pursues wider government access to Mythos, which has demonstrated an extraordinary capability for finding vulnerabilities in computer networks. Its release has been limited by the company for now to a handful of financial institutions and technology companies to test their networks. The White House opposed a recent proposal by Anthropic to expand that group. So far, most federal agencies have not been given the ability to use to Mythos, though Gregory Barbaccia, federal chief information officer of the White House Office of Management and Budget, said in an email that OMB is setting up protections that could allow their agencies to begin using the closely guarded AI model. Last week, Trump signaled that the US would ultimately have a good relationship with Anthropic, a significant shift in tone after months of tensions with the company. In a CNBC interview, Trump said US officials had "very good talks" with Anthropic executives, adding "I think we'll get along with them just fine."

Trump administration blocks Anthropic's Mythos rollout

The Trump administration has told Anthropic it disagrees with a controlled rollout that would bring its dangerous cyberattack-capable AI to roughly 120 organisations in total, even as the White House simultaneously explores an executive order to bring Anthropic back into federal government use. The White House has told Anthropic it opposes the company's plan to expand access to Mythos, its advanced cybersecurity AI model, to roughly 70 additional companies, according to Bloomberg, which cited an administration official speaking anonymously. The development was first reported by the Wall Street Journal. Anthropic declined to comment. Mythos, announced in early April through Anthropic's Project Glasswing initiative, is a model capable of autonomously finding and exploiting vulnerabilities across a wide range of critical software, a capability Anthropic deemed too dangerous for general release. The company has instead been allowing a limited set of organisations to test it on their own systems. Its plan to expand that initial group from approximately 50 to 120 organisations has drawn a direct objection from the Trump administration on two grounds. First, security concerns about the potential for misuse, and a more operational worry that Anthropic does not have enough computing power to serve 120 entities without degrading the government's own ability to use the model effectively. The compute concern is not incidental. Part of the motivation behind Anthropic's current $900 billion fundraising consideration is specifically to secure sufficient infrastructure to run Mythos at scale. The National Security Agency is among the government agencies currently using Mythos, and the White House's concern appears to be that broadening the rollout competes with that access. The White House objection arrives in the context of an already fraught rollout. On the same day Anthropic announced its limited release plan, a small group of unauthorised users in a private online forum gained access to Mythos. The breach, the details of which remain unclear, underlined the gap between Anthropic's controlled-access design and the practical difficulty of containing access to a model of this capability, and has intensified government anxiety about any further expansion of the user base. The model's capabilities are not in dispute. Mythos Preview autonomously discovered thousands of zero-day vulnerabilities across every major operating system and web browser in testing, succeeded on 73% of expert-level capture-the-flag cybersecurity tasks, and became the first model to complete a 32-step simulated corporate network attack end-to-end. These are not hypothetical risks. They are documented capabilities that the US government has both cited as a security threat and apparently sought to monopolise for its own use through the NSA's access. What makes Wednesday's objection striking is its juxtaposition with a separate, simultaneous development; the White House is also currently developing an executive action that would allow government agencies to work around the Pentagon's supply chain risk designation of Anthropic and onboard its models, including Mythos. The Pentagon designated Anthropic an unprecedented national security supply chain risk in early 2026, following a breakdown in negotiations over whether the US military could use Claude for autonomous weapons and domestic mass surveillance, two uses Anthropic's CEO Dario Amodei has publicly said he will not permit. The White House is simultaneously convening companies across sectors this week to inform that draft executive action, including 'table reads' of possible guidance, while also telling Anthropic it opposes its Mythos expansion plan. White House chief of staff Susie Wiles and Treasury Secretary Scott Bessent met with Anthropic CEO Dario Amodei earlier this month in what both sides described as a productive introductory meeting. The official White House line, that it is 'balancing innovation and security while cooperating with the private sector', does not resolve the apparent contradiction between those two tracks. The Mythos dispute is the most visible flashpoint in a genuinely novel regulatory situation: a company whose AI model is simultaneously used by the NSA, opposed by the Pentagon, courted by the White House for re-integration, and now blocked from expanding access to the civilian companies it had approved. The outcome of this week's conversations will shape not just Anthropic's rollout plans but the broader question of how the US government intends to govern AI models capable of offensive cybersecurity operations.

White House against Anthropic expanding Mythos model access: report

San Francisco (United States) (AFP) - The White House is opposing Anthropic's plans to expand access to its new artificial intelligence model Mythos to 120 companies, the Wall Street Journal reported Wednesday. President Donald Trump's administration and Anthropic had only recently started to mend ties following a dispute over the AI firm's refusal to grant the military unconditional use of its software. Anthropic has withheld the powerful Mythos model from public release citing potential cybersecurity risks and concerns it could be exploited by hackers. Instead it shared a version with selected companies including Apple, Microsoft and Nvidia under a project called "Glasswing" to help improve their security infrastructure. The AI startup proposed expanding access to the model to some 70 additional companies, which would bring the total number of organizations with access to around 120, the Wall Street Journal reported citing people familiar with the matter. The White House, which has been at loggerheads with Anthropic for months, opposed the expansion over security concerns, according to the Journal. Authorities were also reportedly worried that Anthropic does not have sufficient computing power to share the technology with the additional companies without hindering the government's ability to use it. In February, Trump instructed the US government to "immediately cease" using Anthropic's technology after Pentagon chief Pete Hegseth designated Anthropic a national security supply chain risk. The company behind the Claude chatbot is now fighting these measures in court. Tensions eased after Anthropic CEO Dario Amodei met with US officials at the White House this month for discussions, which the company described as "productive." Earlier this week, Anthropic said it was investigating unauthorized access to Mythos after Bloomberg News reported that a small group of users in a private online forum had gained access. The California-based developer says Mythos can spot undiscovered security loopholes that have existed for decades, in systems tested by both human experts and automated tools. But the company has also been accused of overhyping the powers of a technology that is its stock in trade, and the subject of neck-and-neck competition with rival OpenAI.

White House Drafts Sweeping AI Guidelines Amid Anthropic Tensions

White House officials are working on developing an artificial intelligence policy that will outline requirements for AI usage across national security agencies. The memo, which touches on the ongoing dispute between the Pentagon and Anthropic, encourages U.S. agencies to use multiple AI providers to avoid relying on a single model, sources told Bloomberg. The policy also notes that AI companies in contract with the Department of Defense must agree not to get involved in the military's chain of command, in which the president is the final decision maker. It was previously reported that the Trump administration is developing a strategy to bypass Anthropic's supply chain risk designation, potentially allowing the onboarding of its powerful AI model, Mythos. Earlier this month, White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent held a meeting with Anthropic CEO Dario Amodei. The meeting was described as a "productive introductory meeting" on potential collaborations between the government and the company. Following the meeting, President Donald Trump said the artificial intelligence company was improving its standing with his administration, raising the prospect that the Pentagon could revisit its ban. The memo is designed to replace the Biden administration's AI policy rather than target specific entities like Anthropic or the Pentagon. It is understood to cover a broad range of topics and is subject to ongoing changes, sources told Bloomberg. The document notes that defense and intelligence agencies add language that protects constitutional rights and prohibits unauthorized surveillance. The memo also establishes a requirement for yearly updates to the Pentagon's autonomous weapon guidelines. Photo: White House in Washington DC under blue sky | Photo Market News and Data brought to you by Benzinga APIs To add Benzinga News as your preferred source on Google, click here.

White House Crafts Plan to Bring Anthropic Back Into Government Fold | PYMNTS.com

The administration is drafting an execution action that could reverse the Office of Management and Budget's directive that limits the use of Anthropic products in the government, according to the report. The White House told Axios, per the report, that it works with frontier AI labs and other organizations to benefit the economy and the country. "However, any policy announcement will come directly from the President and anything else is pure speculation," the White House said, according to the report. Currently, since the White House imposed restrictions on the use of Anthropic products, government agencies can use the company's models while a legal battle between Anthropic and the Pentagon continues, according to the report. However, cooperation between the government and the company is complicated, per the report. The White House told federal agencies in late February to stop using Anthropic's AI products, marking a sharp escalation in a dispute that started inside the Defense Department but expanded to touch the broader government. President Donald Trump said the federal government would no longer work with Anthropic and that agencies using Anthropic's Claude models would get a six-month phaseout period. The decision came just ahead of a Pentagon deadline for Anthropic to agree that the military can use the company's models in "all lawful use cases," a concession the company refused. Anthropic wanted contract language that would prohibit use of its models for autonomous weapons and mass domestic surveillance. Soon after the White House announced this decision, Anthropic said it was planning legal action to challenge the government's action in court. On April 17, it was reported that Anthropic CEO Dario Amodei was meeting with White House Chief of Staff Susie Wiles in the West Wing to resolve the dispute. By that time, Anthropic's Mythos model was causing widespread concern about cybersecurity. Anthropic had been briefing government officials about Mythos and was in talks to give government agencies early access to the model.

White House fights Anthropic's plan to expand Mythos tool that experts fear could cause AI doomsday

The White House is reportedly fighting Anthropic's plan to expand access to Claude Mythos - a powerful AI tool that company execs have warned could cause a wave of hacks and terror attacks if it fell into the wrong hands. Anthropic recently proposed giving an additional 70 companies access to Mythos, bringing the total number to 120 organizations, sources familiar with the matter told the Wall Street Journal. Just earlier this month, the firm announced "Project Glasswing," a plan to provide the model to a select group of handpicked companies including Amazon, Google and JPMorgan. White House officials have told Anthropic that they are against the move to broaden the rollout because of security concerns, sources said. A nightmarish analysis from Anthropic itself earlier showed that Mythos could easily exploit electric grids, power plants and hospitals if hacked. Some Trump administration officials are also reportedly concerned that Anthropic does not have enough computing power to serve both government agencies and the additional companies. A White House official told The Post that the Trump administration is actively engaging with the private sector while trying to balance innovation and security. Anthropic did not immediately respond to The Post's request for comment. Talks between White House officials and Anthropic execs over the Mythos rollout are reportedly seen as an attempt to repair their relationship, as the two are currently entangled in a legal battle working its way through the court in two separate cases. Earlier this year, the Pentagon scrapped its contract with Anthropic and threatened to blacklist the company after it refused to give the government unchecked access to its AI tools, seeking restrictions on their use for mass surveillance or weaponry. Anthropic CEO Dario Amodei met with Susie Wiles, the White House chief of staff, and other senior officials on April 17. Certain government agencies already have access to Mythos, and the Trump administration is working to expand those permissions. Last week, Anthropic said it is investigating potentially unauthorized access to Mythos - intensifying fears around the AI doomsday that could break out if the tool was fully unleashed. Bloomberg reported that a handful of users were able to hack into Mythos on April 8, the same day that Anthropic revealed it was only making the tool available to handpicked corporate clients. Since gaining access, the hackers have been using Mythos "regularly," but not for cybersecurity purposes, according to the news outlet. The AI tool is getting so good at sniffing out and exploiting cybersecurity bugs that Anthropic has been giving security researchers early access to the platform as a preventative measure. Anthropic has recently struck several deals with firms including Amazon, Google and Broadcom to increase its access to computing power over concerns that limited compute could restrict its growth.