US Export Control Directive Suspends Anthropic Fable 5 and Mythos 5 Access Worldwide

The directive compelled immediate global suspension of both models to block foreign national access, including internal Anthropic staff. Anthropic confirmed compliance within hours and stated that other models remain unaffected. The order cited national security authorities without detailing specific threats beyond a single demonstrated bypass technique.

Anthropic's internal review found the reported jailbreak elicited only minor, previously known vulnerabilities already reproducible in GPT-5.5 and other public models. Red-team data from UK AISI and third parties showed Fable 5 safeguards outperformed prior deployments on cyber misuse tasks. No universal jailbreak was demonstrated, and disclosed cases produced no Mythos-specific capability uplift.

This action extends documented export control patterns applied to semiconductor equipment and dual-use software since 2022. The intervention targets deployment rather than weights, indicating direct state oversight of frontier model inference for non-proliferation objectives. Similar restrictions on specific capability thresholds are likely to recur as model releases accelerate.

Anthropic must now implement 30-day data retention across remaining models while monitoring for further directives. Regulators may require pre-deployment reviews for any model exceeding current cyber benchmarks within the next quarter.