Anthropic PBC has disabled access to its advanced artificial intelligence models, including Mythos, for all foreign nationals following a U.S. government order. The directive, confirmed by a U.S. official as originating from the Commerce Department, requires Anthropic to suspend access to its Fable 5 and Mythos 5 models for any foreign national, regardless of their location, citing national security concerns.
Sweeping Government Action
The company has since halted access to both systems for all customers to ensure compliance with the order. This action marks an unprecedented move by the U.S. government to restrict foreign access to frontier AI models developed by an American company. While previous administrations have limited foreign access to technologies like semiconductors and supercomputers, restrictions on AI software itself have sparked debate regarding constitutional and commercial implications.
Anthropic stated that the government’s order appears to stem from the discovery that its Fable 5 model, a recent iteration of Mythos, could be “jailbroken” or have its security guardrails bypassed. The company specifically noted that it had already blocked Fable 5 from performing cybersecurity tasks.
“We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,” Anthropic said in a statement. “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
Concerns Over Vulnerabilities and Precedent
Researchers from Amazon.com Inc. had reportedly identified vulnerabilities in Anthropic’s model through jailbreak research, according to a Wall Street Journal report. People familiar with the matter indicated that Amazon and the U.S. government were in communication regarding these vulnerabilities before the controls were implemented, with Amazon CEO Andy Jassy involved in these discussions. The Information had previously reported that Jassy had raised concerns with senior U.S. officials.
An Amazon spokesperson acknowledged that government consultations on security risks are not uncommon but declined to provide specifics. The government’s broad restriction on AI model access for national security reasons could set a precedent for other major AI developers like OpenAI, Google, and Meta Platforms Inc.
Industry leaders such as Nvidia CEO Jensen Huang and OpenAI CEO Sam Altman have previously advocated for promoting the global adoption of American AI systems and maintaining the nation’s technological lead. Aidan Gomez, co-founder of Cohere Inc., described the situation as a “massive wake-up call” for those who might have underestimated the potential for such leverage to be exerted.
Background and Industry Response
Anthropic had previously secured a court order against a Trump administration ban on government use of its AI technology, arguing that such a ban could result in billions of dollars in lost revenue. The company received the latest government order on Friday at 5:21 p.m. New York time. This directive contrasts with earlier statements and an executive order from the Trump administration that suggested a move away from a mandatory licensing regime for model reviews.
The current restriction also appears to build on prior tensions between Anthropic and elements within the Trump administration, including a dispute over the use of its technology for military and surveillance purposes. The administration had previously declared the company a U.S. supply-chain risk and ordered U.S. agencies to phase out its products.
Privately held Anthropic, which has positioned itself as a responsible AI developer, initially released its Mythos model in April to a limited group, citing risks associated with its cybersecurity vulnerability detection capabilities. Despite these concerns, signs had emerged that the limited release was easing tensions, with reports indicating the U.S. government was preparing to make a version of Mythos available to federal agencies.
David Sacks, a former AI czar for Trump, stated that Anthropic had refused to fix a jailbreak issue in its Fable model. He expressed the administration’s hope that Anthropic would remediate the safety issue, leading to the lifting of export controls and a general release of Fable. Sacks added that the administration is bewildered by Anthropic’s apparent reluctance to comply with safety requests.
The new government restriction comes amid an intense race among U.S. developers to release advanced AI models and demonstrate profitability to investors. Both OpenAI and Anthropic are reportedly pursuing initial public offerings this year.
Anthropic itself had recently called for a system where governments and AI developers collaboratively decide on slowing down work on potentially dangerous AI technologies. The company noted that AI is advancing to a point where it could significantly increase human efficiency or even replace human work, creating new risks.
The European Union’s executive arm is assessing Anthropic’s statement and consulting with allies on the potential risks and cybersecurity concerns related to powerful new AI models, underscoring Europe’s need for technological sovereignty.
Helene Elliott is the senior reporter for News Raise. She covers Science news. She also has a keen interest in photojournalism. Helene holds a nomination for the prestigious Red Smith Award. She is married to author Dennis D’Agostino, a former publicist with the New York Mets.
