Anthropic AI Models Offline Amidst Clash with Trump Administration
Anthropic's advanced AI models, Mythos and Fable, have been taken offline following stringent export controls imposed by the Trump administration. Administration officials claim Anthropic failed to comply with a cyber executive order and distributed models susceptible to 'jailbreaking.' This action stems from alleged communication failures and concerns, including earlier reports of the Mythos model's availability to an entity with ties to the Chinese Communist Party.

Anthropic's powerful artificial intelligence models, Mythos and Fable, have been removed from the internet after the Trump administration imposed stringent export controls.
Administration officials allege that Anthropic did not 'honor' a recent cyber executive order and failed to adequately address concerns regarding the security of its products. An administration official stated that Anthropic was perceived as a 'bad actor' and questioned giving the company a chance, claiming they 'screwed us.'
On Thursday, Amazon CEO Andy Jassy reportedly contacted Treasury Secretary Scott Bessent, expressing concerns that Anthropic's Mythos and Fable models could be 'jailbroken.' An administration official suggested Anthropic was aware of this vulnerability but proceeded with distribution. Conversely, Anthropic maintains it received explicit government approval to deploy Fable.
The White House previously threatened Anthropic with export controls weeks prior, after learning its Mythos model was accessible to a foreign entity with direct ties to the Chinese Communist Party. A source close to Anthropic stated the company routinely collaborates with the government on Mythos access and proactively revoked access in the aforementioned case, involving a global telecom company, without the threat of export controls. While Amazon's report raised new concerns, an administration source indicated Anthropic's initial stance was that it was 'not a real issue,' a claim Anthropic disputes, stating it did not refuse to resolve the issue.
The White House noted that Anthropic's communication challenges have been a recurring theme, citing a past disagreement with the Pentagon, although officials described that incident as unrelated to the current situation. Officials emphasized a desire for innovation but stated their 'hands were tied.'
In response to the Amazon report, Anthropic issued a blog post dismissing its findings. The company then engaged a cybersecurity expert, whom the administration reportedly viewed as a 'radical Democrat,' and who was subsequently acknowledged by Chris Krebs, recently terminated by President Trump.
Meetings are scheduled for Monday between the Commerce Department and Anthropic senior tech staffers Logan Graham, Dave Orr, and Nicholas Carlini. Additional meetings are planned with the CIA and White House science advisor Michael Kratsios to address compliance with the cyber executive order. Addressing jailbreak vulnerabilities remains a key objective, though achieving perfect resistance may be challenging. According to Axios, an administration source suggested that an 'attitude fix' might be necessary to ensure all parties feel 'safe, secure and happy.'


