White House forces Anthropic to pull Fable 5 AI model after Amazon security report

White House forces Anthropic to pull Fable 5 AI model after Amazon security report

Anthropic's Fable 5 AI model was removed from public access Friday night after Amazon reported security vulnerabilities to the White House, triggering export controls that made the model inaccessible to foreign nationals and U.S. allies.

What happened

Amazon contacted White House officials Thursday night with a report showing they had jailbroken portions of Anthropic's Mythos model that "pose a national security threat," according to sources. Fable 5 is the general-use version of Mythos.

At least five other companies also contacted senior administration officials between Thursday evening and Friday morning about the model, sources told Axios.

Anthropic had notified the government multiple times about the planned June 9 release and received no objections, according to a source close to the company.

By Friday 5:20 p.m. ET, the White House sent Anthropic a letter imposing export control rules on Fable and Mythos models. Users lost access by approximately 10 p.m.

The technical dispute

Security expert Katie Moussouris of Luta Security, who reviewed Amazon's report, said the government's response "seems way out of line with what's actually in the research report."

Moussouris said researchers found vulnerabilities by asking questions normal security defenders would ask AI — exactly what the model was designed to do. "All AI models need to be able to help defenders in exactly this way, or we won't be able to scale our defense against attackers," she said.

Anthropic officials told the administration the alleged jailbreak was "relatively simple, could be achieved with other models, and did not demonstrate a flaw in Fable 5's safety systems," according to sources.

Amazon's role

The intervention raises questions about why Amazon would act against a company in which it is a major investor. Amazon has made substantial investments in Anthropic for computing infrastructure.

An Amazon spokesperson said: "As a leading cloud provider that serves a large number of private and public sector customers, it's not uncommon for governments to seek our counsel on potential security risks. When they occur, we don't share the details of these discussions."

Impact on industry

The export controls prevent access not only to foreign adversaries — which Anthropic was already blocking — but also to U.S. allies and foreign nationals working in the U.S. This immediately affected Anthropic employees, many of whom are foreign-born and need model access for their work.

One person familiar with the situation called it "a de-facto licensing regime," adding: "Companies will not screw with the White House. That is the ultimate effect."

An administration official said other models are not viewed as national security threats because they don't surpass Mythos's capabilities. Any model at Mythos level or above would need administration approval to ensure "the government's national security apparatus is hardened enough."

What this means

This marks the first time the U.S. government has used export controls to force a model takedown post-release. The incident reveals a reactive regulatory approach where competitors' security reports can trigger immediate government intervention, even when the AI company previously coordinated with authorities. The precedent suggests companies building frontier models may face unpredictable regulatory risk based on third-party assessments, regardless of prior government clearance. The technical merit of Amazon's security findings remains disputed by independent experts.