Anthropic has disabled all access to its Fable 5 and Mythos 5 artificial intelligence models following a sudden export-control directive from the United States government.
Issued at 5:21 PM ET on June 13, 2026, the directive cited pressing national security concerns and strictly prohibited any foreign national from accessing the models.
This restriction extends beyond public users, legally barring even foreign-national employees at Anthropic from interacting with the systems. To ensure strict regulatory compliance, Anthropic disabled both models worldwide, though all other legacy models remain fully operational.
The government’s unprecedented intervention reportedly stems from a newly discovered method to bypass or “jailbreak” Fable 5’s safety guardrails.
Anthropic Blocks Fable 5 and Mythos 5
While the directive lacked specific technical documentation, Anthropic stated that the government provided verbal evidence of a narrow, non-universal jailbreak.
The exploit specifically involves manipulating the model’s prompts to force it to review a targeted codebase and autonomously identify or remediate software flaws.
Following an internal review of the technique, Anthropic disclosed that the bypass only successfully identified a small subset of minor, previously documented vulnerabilities.
The company vehemently argued that this exploit provides no unique capability uplift or zero-day discovery advantage.
According to Anthropic, threat actors could easily discover these exact vulnerabilities using publicly available models. The company explicitly cited OpenAI’s recent disclosures of GPT-5.5 cybersecurity evaluations as evidence of comparable industry capabilities.
Anthropic is publicly disputing the severity of the threat, pointing to the extensive security evaluations Fable 5 underwent prior to deployment.
The model was subjected to thousands of hours of rigorous stress testing by internal red teams, the UK AI Safety Institute (AISI), private third-party security researchers, and the US government itself.
These comprehensive evaluations confirmed that Fable 5’s safety mechanisms significantly outperform previously deployed frontier models. Testers entirely failed to discover a universal jailbreak that could broadly bypass the model’s safeguards across various cyber-attack vectors.
Because perfect jailbreak resistance remains practically impossible, Anthropic designed Fable 5 with a strict defense-in-depth architecture.
This security framework relies on several core mitigation strategies, primarily confining non-universal jailbreaks to an exceptionally narrow operational scope and structuring technical safeguards to make developing universal jailbreaks economically prohibitive.
To support this, Anthropic deploys continuous telemetry monitoring to rapidly detect anomalous prompt behavior and shut down active attacks.
The company also enforces a 30-day customer data retention policy for Mythos-class models, which is essential for facilitating continuous threat hunting and patch development.
While complying under protest, Anthropic warned that applying this extreme recall standard could have a chilling effect on the technology sector.
The company argued that halting a deployment over entirely benign responses or minor, non-unique findings would effectively freeze all future model releases for every major AI provider.
Anthropic states that while governments must retain the authority to block demonstrably unsafe AI, such actions must be grounded in transparent, fair, and technically accurate criteria.
The company has apologized to its enterprise customers for the immediate operational disruption and is actively petitioning to restore access. Security researchers can expect Anthropic to release a detailed technical disclosure regarding the jailbreak mechanics within 24 hours.
Follow us on Google News, LinkedIn, and X to Get Instant Updates and Set GBH as a Preferred Source in Google.
