The White House is reportedly considering reinstating Anthropic as an approved technology provider across federal agencies, despite ongoing disputes with the Pentagon regarding usage of its AI models. Axios, citing multiple sources, reports that officials in President Donald Trump’s administration are preparing guidance to bypass Anthropic’s designation as a ‘supply chain risk,’ thus allowing access to its latest AI model, Mythos.
This potential policy shift represents a significant change from the previous stance where Anthropic was viewed as a security threat. This followed CEO Dario Amodei’s refusal to provide unrestricted Pentagon access earlier this year. The administration had initially removed Anthropic’s technology from government systems after designating it a supply chain risk in February, with Trump urging agencies to stop using its products immediately.
A draft executive action under discussion could soften the administration’s stance, allowing federal agencies to utilize Anthropic’s technology while maintaining political leverage. In his February statement on Truth Social, Trump emphasized that military operations should not be influenced by what he described as a ‘radical left, woke’ company: “The United States of America will never allow a radical left, woke company to dictate how our great military fights and wins wars!”
Anthropic’s CEO Dario Amodei reportedly engaged with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent earlier this month. Concurrently, administration officials met with major banks such as Citigroup, Bank of America, Wells Fargo, Morgan Stanley, and Goldman Sachs to address cybersecurity risks associated with Mythos.
Interest in Mythos is growing among federal agencies and corporations since its reveal in March as Anthropic’s most advanced AI model capable of identifying software vulnerabilities across major operating systems. Anthropic has restricted public release but granted access to tech giants and governments for security testing. The AI Security Institute noted that Mythos Preview completed ‘The Last Ones’ (TLO) challenge, a 32-step corporate network attack simulation.
The National Security Agency reportedly uses Claude Mythos Preview in classified settings, while Mozilla used it to discover and fix 271 browser vulnerabilities during tests. Mozilla highlighted the model’s potential for both exploiting and protecting software from threats: “As these capabilities reach more defenders, many teams are experiencing the same vertigo we did when findings first came into focus,” wrote Mozilla.
Anthropic has not yet responded to Decrypt’s request for comment on the matter.