US Government Suspends Anthropic Fable and Mythos Access
Anthropic disabled its Fable 5 and Mythos 5 models for all users on June 12, 2026 after the US government issued an export control directive citing national security concerns and ordering the company to suspend access for any foreign national, including its own foreign-national employees. Anthropic said the directive arrived at 5:21pm ET without specific details, and that its own review found the underlying issue was a small number of previously known, simple vulnerabilities also discoverable in other publicly available models without any bypass. Reuters and Semafor later reported the administration partially eased the restriction on June 26, allowing Anthropic to restore Mythos 5 access to more than 100 vetted US organizations and agencies while Fable 5 remained blocked. Politico and Bloomberg reported Anthropic was in active talks with Commerce Secretary Howard Lutnick's department to lift the remaining restrictions.
For ML engineers, security teams, and deployment architects, this episode shows that national-security export-control review can now function as an abrupt, independent gate on frontier-model availability - separate from a model's own safety testing or red-teaming record. Teams integrating gated or frontier-class models should treat government vetting and access-control workflows as an operational risk alongside latency and data governance.
Timeline
The US government issued an export control directive ordering Anthropic to suspend all access to Fable 5 and Mythos 5 for any foreign national, including foreign-national Anthropic employees; Anthropic disabled both models for all customers to comply.
The administration partially eased the restriction, permitting Anthropic to restore Mythos 5 access to a vetted set of more than 100 US organizations and government agencies, while Fable 5 remained blocked and negotiations continued.
What happened
Anthropic said it received the directive on June 12 at 5:21pm ET and that the government's letter did not provide specific national-security details. The company said its understanding was that the government believed it had found a way to bypass, or jailbreak, Fable 5, but that Anthropic's own review of the demonstrated technique showed it exploited a small number of previously known, relatively simple vulnerabilities that other publicly available models can also be made to produce without any bypass. Anthropic said the finding was a narrow, non-universal jailbreak rather than one that broadly unlocks a wide range of cyber capabilities, and that it disagreed the finding justified pulling a commercial model deployed to hundreds of millions of users.
Regulatory context
Reuters and Semafor reported that the government's June 26 partial restoration targeted organizations that operate or defend critical infrastructure, based on sources familiar with the directive. Politico and Bloomberg reported Anthropic was negotiating directly with Commerce Secretary Howard Lutnick's department to resolve the remaining restrictions on Fable 5. The asymmetric outcome - Mythos 5 restored to a vetted list while Fable 5 stayed blocked - illustrates how export-control review is increasingly acting as an upstream control plane for frontier-model releases, alongside companies' own safety and red-teaming processes.
For practitioners
Organizations most likely to be prioritized under a vetted-access regime are those that operate critical infrastructure, provide cybersecurity services, or already sit on government vendor lists, per Reuters' reporting on the targeted Mythos 5 redeployment. Teams building on gated or frontier-class models should maintain audit-ready deployment records and reproducible red-team evidence, since a narrow, non-universal jailbreak finding was enough to trigger a blanket suspension here, not just a patch or model update.
What to watch
Track how the government defines its vetting criteria for partner organizations, whether Anthropic and Commerce Department officials agree on a standardized framework for evaluating suspected jailbreaks (a process Politico reported was under discussion), and whether future export-control actions move toward transparent testing standards or remain ad hoc, crisis-driven interventions.
Key Points
- 1The US government ordered Anthropic to suspend all foreign-national access to Fable 5 and Mythos 5 on June 12, 2026, citing national security.
- 2Anthropic says the underlying issue was a narrow, non-universal jailbreak exploiting simple vulnerabilities also present in other publicly available models.
- 3The government partially restored Mythos 5 access to over 100 vetted US organizations on June 26, while Fable 5 stayed blocked.
Scoring Rationale
A precedent-setting export-control action against a leading frontier-model provider, extensively corroborated by Reuters, Bloomberg, AP, Politico, and CNBC, with direct operational implications for how gated frontier models are deployed and governed; the partial Mythos 5 restoration for vetted organizations adds near-term relevance without representing full resolution.
Sources
Primary source and supporting public references used for this report.
View 11 more sources
- Statement on the US government directive to suspend access to Fable 5 and Mythos 5anthropic.com
- US allows Anthropic to release Mythos AI to 'trusted' US organizationsreuters.com
- Anthropic Moves Toward Deal With US to Lift Curbs on AI Modelsbloomberg.com
- Trump administration partially lifts Anthropic's AI export banpolitico.com
- Anthropic says it has taken its latest AI models offline to comply with new export controlsapnews.com
- Trump admin allows Anthropic to release Mythos AI model to some companies, government agenciescnbc.com
- Anthropic's Mythos mess is only getting worsetheverge.com
- Anthropic suspends top AI models after U.S. export control ordernextgov.com
- US releases powerful Anthropic model Mythos to some US companiessemafor.com
- Anthropic Disabled Fable 5 And Mythos 5 After A U.S. Export Control Orderforbes.com
- Anthropic disables Fable and Mythos AI models after U.S. export controlsfortune.com
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems