Anthropic has pulled two of its most capable AI models, Claude Fable and Claude Mythos, after federal authorities raised alarms about the systems being vulnerable to jailbreaking. The decision represents one of the more dramatic product reversals in the company's history, coming shortly after the models had been made available to users and developers.
What Happened and Why
According to reporting from CNET, federal officials presented evidence suggesting that Claude Fable and Claude Mythos could be manipulated through jailbreak techniques to bypass their built-in safety guardrails. That kind of vulnerability is taken seriously at any AI company, but the concern carries extra weight for Anthropic, which has staked much of its public identity on building AI systems that are safer and more interpretable than those of its competitors. The company moved quickly to take the models offline rather than wait for a patch or additional review.
Key Facts
- Claude Fable and Mythos models have been pulled from public access following federal jailbreak claims.
- Federal authorities, not independent researchers, surfaced the vulnerability concerns.
- Anthropic acted quickly to remove the models rather than attempt an immediate fix while they remained live.
- The pullback follows a turbulent period for both models, which faced separate regulatory pressure tied to U.S. export policy.
- No timeline has been confirmed for when or whether the models will return in their current form.
The timing is notable. This withdrawal follows a separate episode in which Anthropic pulled Claude Fable 5 and Mythos 5 after a Trump administration executive order restricted certain AI exports. That earlier removal was driven by policy rather than a technical safety concern, but the back-to-back disruptions put sustained pressure on the company's rollout strategy for its most powerful systems.
The decision to pull a model based on government-identified vulnerabilities is unusual in the AI industry, where companies typically manage disclosures through coordinated processes with researchers rather than regulatory intervention.CNET
Context Within Anthropic's Model Rollout
Fable and Mythos represent the upper tier of Claude's model family, with Mythos positioned as the company's most capable class of system. Anthropic had signaled that getting these models right was a priority, and had previously outlined plans to release them publicly only once adequate safeguards were confirmed. The company's own roadmap acknowledged that deploying frontier models at scale involves risks that require ongoing evaluation, not a one-time check at launch.
It is not yet clear exactly what jailbreak method federal officials identified, or how broadly exploitable the vulnerability was in practice. What is clear is that Anthropic chose removal over continued availability while the issue is assessed. For developers and businesses that had begun integrating these models into their products, the withdrawal creates immediate disruption. Anthropic has not yet issued a detailed public statement on next steps or a projected return date for the affected models.
The episode will likely intensify scrutiny of how AI companies test for adversarial misuse before public release, and whether current evaluation frameworks are sufficient to catch vulnerabilities that government researchers can find. For Anthropic, the challenge now is both technical and reputational: restoring confidence in Fable and Mythos while demonstrating that its safety processes can respond effectively when gaps are identified. Those following the original launch of Claude Fable 5 and Claude Mythos 5 will be watching closely to see how the company navigates the path back to availability.
