The US government ordered Anthropic to deactivate its most powerful artificial intelligence models, Fable 5 and Mythos 5, with less than 90 minutes’ notice last Friday. The directive, issued by the White House, cited national security as the justification, but provided no specific details to the company or its 3,000 employees. This abrupt shutdown left Anthropic managers unable to explain the outage to customers and employees questioning the reasons behind the action.
Anthropic employees, whose work on the Fable 5 and Mythos 5 models was halted mid-task, expressed confusion in internal group chats. The lack of official explanation led to speculation among staff. One employee asked, “Are we being bullied based on bad vibes?” Another suggested a darker possibility, querying, “At what point does this just feel like they don’t want us to exist?” The models were the most advanced Anthropic had developed.
The stated reasons for the shutdown shifted multiple times, according to internal discussions among engineers. Initial concerns pointed to foreign companies gaining access to the models. This explanation later evolved to include a potential vulnerability within the models themselves, and then a combination of both. Employees noted that news reports circulating contradicted each other, adding to the uncertainty regarding the official stance.
The government order broadly prohibited any foreign national, including green-card holders working in the US and individuals outside the country, from accessing the models. This wide scope effectively necessitated a complete shutdown of the systems for all users. Anthropic executives immediately contacted officials following the order, seeking clarification on the specific reasons, but received no detailed explanation.
Anthropic independently identified a research paper as the potential trigger for the government’s action. The company then contacted the White House to confirm if this paper was indeed the cause. This indicates the limited communication provided by government agencies regarding the shutdown’s basis.
The research paper originated from Amazon, a company that has committed to investing up to Amazon 33 billion in Anthropic. Amazon researchers demonstrated that the Fable 5 model could be manipulated to identify flaws in vulnerable code. Amazon CEO Andy Jassy reportedly conveyed these concerns directly to Treasury Secretary Scott Besse, which may have prompted the White House’s intervention.
The immediate future for Anthropic’s Fable 5 and Mythos 5 models remains unclear, with no public timeline for their potential reactivation or resolution of the underlying security concerns. The incident raises questions about the transparency of government intervention in advanced AI development and the criteria used to assess national security risks in this rapidly evolving field.
Observers will be watching for further details from both the US government and Anthropic regarding the specific vulnerabilities identified and any new protocols or regulations that may emerge from this event. The incident highlights the ongoing tension between rapid AI innovation and the imperative for national security oversight.