TL;DR
- Warning Path: Amazon CEO Andy Jassy and other tech leaders have warned U.S. officials about Fable 5 risks.
- Access Cutoff: The directive suspended Fable 5 and Mythos 5 for foreign nationals and took both models offline.
- Risk Dispute: Anthropic says the issue was a narrow prompt bypass, while David Sacks says officials wanted fixes.
- User Impact: Affected users remain blocked unless Anthropic submits changes that the government accepts.
On June 12, a White House directive forced Fable 5 and Mythos 5 offline after a phone call from Amazon CEO Andy and warnings from other tech leaders reportedly led officials to assess a Fable 5 threat, people familiar with the matter said. Customers and partners immediately lost access to the powerful two AI models.
After the calls, White House officials reportedly assessed whether the model risk justified a national-security access restriction.
Amazon CEO Andy Jassy appears to have been the source of security concerns that preceded the cutoff. His call and notices from five or more other firms about Fable 5 reportedly happened during the same Thursday evening to Friday morning window.
Foreign nationals, including people who are not U.S. citizens or permanent residents inside the company, lost Fable and Mythos access under the directive. Anthropic temporarily took the affected models offline for all customers, while the rest of the Claude lineup remained available.
What makes the case complicated, is the government’s choice of an export-control directive is a national-security order restricting who can use the models. Amazon has not confirmed the detailed warning call; its public position is that governments seek Amazon’s counsel on security risks and that the company does not disclose those discussions. Anthropic says it disputes the underlying premise of security rists and is working to restore access as soon as possible.
Why the Warning Mattered
Amazon researchers appear to have found a Fable 5 prompt workaround/jailbreak that could produce information useful for cyberattacks. Jassy then raised the security concern with senior officials, including Treasury Secretary Scott Bessent.
Declined White House demands to Anthropic to fix the jailbreak then turned into a cutoff for customers and foreign-national employees. According to Anthropic, the directive reached the company without specific national-security details in the letter, and the affected models were offline within hours.
David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, said the administration had asked Anthropic to fix or remove the model: “The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused.”
I’ve had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true:
— As we know, Anthropic publicly released its Mythos class models earlier this week under the commercial name Fable.…
— David Sacks (@DavidSacks) June 13, 2026
Sacks’s account makes remediation the government’s public condition, not just punishment for a disputed model behavior. Recent federal interest in a reported classified deal for Anthropic AI gives the dispute a policy backdrop beyond just one alleged jailbreak.
Anthropic´s customers now face the question: can Anthropic change Fable 5 enough to satisfy officials without keeping both affected models offline for an extended period.
Anthropic Pushes Back on the Risk
Anthropic argues that the disputed capability is already available in other public models and that the order rests on a non-universal jailbreak, meaning a narrow prompt bypass rather than a broad removal of safeguards. In practical terms, Anthropic’s position is that the government treated a limited model-safety finding as grounds to recall a commercial system used by large numbers of people.
Older jailbreak research helps explain the technical concern without making it a parallel current dispute. In 2024, Anthropic identified many-shot jailbreaking as a way to make large language models answer harmful questions.
Fable 5’s launch days earlier positioned the model around safety routing for sensitive requests.
Cybersecurity expert Katie Moussouris reviewed Amazon material at Anthropic’s request and concluded “I’ve seen the research paper. It’s not a jailbreak. It was Defense Oriented Prompting (DOP) – a capability defenders need.” Sacks denied allegations the action was tied to earlier Anthropic disputes with the Defense Department.
Competing narratives now leave the new conflict focused on the scope of a bypass and the policy fight focused on who decides when a model must come down.
What Comes Next for Fable Access
Fable 5 had entered public availability days earlier as a Mythos-class Claude release with safety routing for sensitive requests.
The short time the model was publicly available has turned into an availability dispute before customers had much time to judge the product.
Amazon’s role is sensitive because the company at the same time is a major Anthropic investor and infrastructure partner, as Jassy’s outreach may have been part of the security-warning .
Fable 5 and Mythos 5 remain offline for affected users unless officials revise the June 12 directive or Anthropic submits fixes the government accepts.
