Following an unprecedented directive from the US government to block all foreign nationals from accessing its technology, Anthropic has shut down access to its most advanced artificial intelligence models, including Mythos.
The company stated in a declaration that the US government required it to suspend all foreign nationals' access to the Fable 5 and Mythos 5 models, "regardless of whether they are physically located in the United States or abroad," citing national security. A US official confirmed that the Department of Commerce issued the letter. Subsequently, the model developer cut off all client access to these two systems to ensure compliance.
The US government has never before taken such a sweeping measure to restrict foreign access to cutting-edge AI models developed by American companies. Both the Trump and Biden administrations have previously restricted exports of other critical technologies like semiconductors and supercomputers, and there has been discussion about potentially blocking access to AI models. However, restrictions targeting the software itself raise multiple constitutional and commercial concerns.
Anthropic stated that the government issued the order because it discovered a potential for "jailbreaking"—or bypassing safety guardrails—in Fable 5, a recently released derivative version of Mythos. The company had already prohibited Fable 5 from performing cybersecurity tasks.
"We believe it is inappropriate to recall a commercial model used by hundreds of millions of users solely due to the discovery of one limited potential jailbreak risk," Anthropic said in a post on its official website. "If this standard were applied industry-wide, the deployment of new models by all leading-edge model providers would essentially grind to a halt."
The government's broad restrictions on a series of AI models in the name of national security could set a precedent for all major AI model developers, including OpenAI, Alphabet Inc. (Google's parent company), and Meta. Industry leaders like NVIDIA CEO Jensen Huang and OpenAI CEO Sam Altman have previously urged the US government to instead promote the global proliferation of American AI systems to maintain the country's competitive advantage.
Anthropic said it received the government order at 5:21 PM New York time on Friday. This directive, issued just before the close of business on Friday, contradicts previous statements from the Trump administration and a recently signed executive order, which indicated the government would not implement a licensing regime for model review.
Friday's order may also intensify long-standing tensions between Anthropic and certain factions within the Trump administration. Earlier this year, the AI developer clashed with the Pentagon over the use of its technology for military and surveillance purposes. Following that conflict, the government listed the company as a supply chain risk and ordered federal agencies to phase out its products.
Privately-held Anthropic, which has long positioned itself as a "more responsible AI developer," initially released the Mythos model to a very limited number of businesses and institutions in April of this year, warning that its capability to discover cybersecurity vulnerabilities made broad distribution too risky.
There were indications that this limited release could have eased tensions between Anthropic and the Trump administration: previous reports suggested the US government was preparing to allow key federal agencies to use a version of Mythos.
Mythos also spurred the Trump administration to accelerate its AI policy formulation, with a recent executive order proposing a voluntary model review system and explicitly stating that the order does not constitute a mandatory licensing regime.
The government's latest restrictions run counter to the trend of US AI developers racing to launch the most advanced models to prove the technology's profitability to investors. Following SpaceX's historic IPO, both OpenAI and Anthropic plan to go public as early as this year.
In a bid to take the lead in releasing frontier AI models, Anthropic published a lengthy blog post earlier this month advocating for a mechanism where the government and AI developers jointly decide when to slow down technological development to mitigate potential risks.
"Globally, it would be best to have an option to demonstrate or temporarily pause" dangerous AI development, the company stated in the blog at the time. AI technology is advancing rapidly, with its efficiency potentially boosting human productivity by thousands of times or even replacing human jobs, giving rise to a series of new risks.
The European Commission stated it is evaluating Anthropic's announcement and continues to communicate with allies regarding potential risks and cybersecurity issues related to advanced AI models. The European Commission added that the latest developments highlight the necessity for Europe to advance its technological autonomy.
Comments