Anthropic presses White House to reverse AI model restrictions
A U.S. directive limiting access to Claude Mythos 5 and Fable 5 has forced Anthropic into urgent talks with Trump administration officials.
By Hana Yoshida · Markets Reporter
3 min read
Anthropic is trying to persuade the Trump administration to lift restrictions that halted broad access to its newest Claude AI models, The Verge reported. The dispute could shape how Washington treats advanced AI systems that can aid cybersecurity work and potentially be used for attacks.
The company received a U.S. export control directive late Friday ordering it to suspend access to Claude Mythos 5 and Fable 5 for any foreign national, including Anthropic employees, according to The Verge. Anthropic concluded that complying with the order required disabling products it had recently promoted.
A person familiar with the talks told The Verge that the administration first called Anthropic on Friday afternoon and gave it 90 minutes to cut off access or face export controls under Commerce Department authority. The person said Anthropic executives contacted the White House within minutes, and CEO Dario Amodei later joined discussions with Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross.
Jailbreak concerns drove the order
Anthropic said Friday that the government appeared to be acting on a reported way to bypass Fable 5’s safeguards. The company described the issue as narrow and non-universal, and said the same level of capability was available from other models, including OpenAI’s GPT-5.5.
The models at issue stem from the same base as Mythos Preview, which Anthropic had previously said was too dangerous for public release, The Verge reported. Mythos 5 was limited to selected government agencies and companies, while Fable 5 included extra safeguards and was cleared by Anthropic for general use.
Semafor reported that U.S. officials were concerned a China-linked group may have accessed the technology. The Verge cited a person familiar with the negotiations saying those concerns dated back weeks and involved a major global telecommunications company that had initially been cleared for Mythos Preview access; Anthropic revoked that access after the government raised concerns, the person said.
David Sacks, the former U.S. AI and crypto czar, wrote on X that a trusted partner of Anthropic and the U.S. government had found a jailbreak while testing Fable. The Wall Street Journal reported that Amazon CEO Andy Jassy raised concerns with officials after Amazon researchers red-teamed Fable 5. The Verge reported that Amazon research came up in government discussions and that Anthropic had been discussing the paper with Amazon researchers before the directive.
Industry warns of wider fallout
Anthropic sent employees to Washington after virtual meetings over the weekend, including safeguards chief Dave Orr, red-team leader Logan Graham and cybersecurity researcher Nicholas Carlini, according to The Verge. Axios reported, citing a person familiar with the administration’s view, that Anthropic had struggled in its communications with the White House.
A public letter from technology and cybersecurity executives urged officials to lift the Fable 5 restrictions. The letter said any AI rules should be based on scientific evaluations developed with industry and academic input.
Alex Stamos, chief product officer at Corridor, told The Verge that he organized the letter because AI systems are needed to find and patch software flaws before attackers do. He said Anthropic had overstated Mythos’ capabilities and warned that companies are signing backup deals with non-U.S. providers because the order exposed political risk.
Ben Van Roo, co-founder and CEO of Legion Intelligence, told The Verge that barring all foreign nationals from using a model would be extremely difficult to enforce. He said the episode raised new questions about who should be allowed to use powerful AI tools as the technology develops faster than regulation.
The Verge reported that OpenAI, Google and Microsoft have released comparable products, which could make the Anthropic order a test case for other AI companies. A person familiar with the talks told The Verge that some administration officials acknowledged export controls on model providers could be hard to apply, especially while the Commerce Department is exploring efforts to promote exports of American AI systems.
This story draws on original reporting from The Verge.