Amazon's Jassy Alerted White House to Anthropic Fable 5 Security Flaws, Triggering Export Ban

Amazon researchers used a series of prompts to get Fable 5 to provide information that was supposed to be off-limits — specifically, data that could aid cyberattacks, according to the WSJ report ^[1]. The underlying jailbreak technique involved asking the model to read a specific codebase and identify software flaws, effectively turning Fable 5's advanced reasoning capabilities into a vulnerability-discovery tool ^[2].

Jassy brought the findings to Bessent and other U.S. officials. The White House convened a meeting to discuss response options, and security researchers tested Amazon's claims before President Trump approved the foreign access ban ^[1]. The administration had 'long felt that Anthropic couldn't be trusted to manage the security risks its new model presented,' the WSJ reported ^[1].

Amazon issued a measured statement acknowledging the contact with government officials. 'It is not uncommon for governments to seek our counsel on potential security risks,' the company said, declining to provide specifics about the discussions ^[4].

Commerce Secretary Howard Lutnick's letter to Amodei directed Anthropic to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States ^[2]. The directive applied to Anthropic's own workforce, creating an immediate operational challenge for the company, which employs foreign-born engineers and researchers.

Anthropic chose to disable both models for all users globally rather than attempt to selectively filter foreign nationals from domestic users in real time ^[2]. The shutdown affected enterprise customers including banks and government agencies that had been using Mythos-class models for complex reasoning and vulnerability discovery ^[4].

David Sacks, the former White House AI and crypto czar who now co-chairs the President's Council of Advisors on Science and Technology, said a 'highly credible trusted partner of both Anthropic and the USG came forward with a jailbreak.' Sacks added that officials requested Anthropic either fix the vulnerability or remove the model, but 'Dario refused' ^[3].

Anthropic pushed back sharply against the government's characterization. In a public statement, the company said it had reviewed a demonstration of the jailbreak and determined it exposed only 'a small number of previously known, minor vulnerabilities' ^[2]. The company argued that the capabilities flagged by Amazon already exist in other publicly accessible AI models.

'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 wrote ^[2]. The company noted that 'perfect jailbreak resistance is not currently possible' and described its approach as a 'defense in depth strategy' combining narrow jailbreak resistance with monitoring capabilities.

Despite its disagreement, Anthropic confirmed compliance with the directive. 'We believe this is a misunderstanding and are working to restore access as soon as possible,' the company said ^[2]. All other Anthropic models, including older Claude variants, remain unaffected by the order.

The situation highlights an unusual tension in the AI industry's tightly interwoven investment and competitive relationships. Amazon is simultaneously Anthropic's largest financial backer and, through AWS, its primary infrastructure provider. In April 2026, Anthropic committed to spending over $100 billion on AWS infrastructure, and Amazon added a fresh $5 billion investment on top of its previous $8 billion stake ^[3].

Amazon also competes directly with Anthropic through its own Nova family of AI models, creating a potential conflict of interest that industry observers have noted. The company that bankrolls Anthropic's infrastructure is the same one that told the government Anthropic's models are dangerous ^[4].

Amazon shares closed at $238.55 on Friday, down 1.2% on the session and off roughly 11% over the past month ^[5]. The stock trades well below its 52-week high of $278.56.

Anthropic said it is working with the government to resolve what it characterized as a misunderstanding and restore access to Fable 5 and Mythos 5 ^[2]. The timeline for any restoration remains unclear, and the directive's scope — covering all foreign nationals regardless of location — sets a precedent for how the administration may handle future AI model security concerns.

The shutdown arrives at a sensitive moment for Anthropic, which had been widely expected to pursue an initial public offering later in 2026 ^[3]. A prolonged ban on its flagship models would complicate that timeline and raise questions about the regulatory risk premium for frontier AI companies.

Mythos-class models were designed for complex reasoning and long-horizon agentic work, and are actively used by financial institutions and government agencies for vulnerability discovery ^[4]. Their absence leaves a gap that competitors — including Amazon's own AI division — stand to fill.