US Issues Export Control Directive on Fable 5 at 5:21pm ET Citing Jailbreak

Anthropic stated the directive arrived at 5:21pm ET and required immediate compliance to block all users. The company described the demonstrated bypass as narrow, limited to rediscovery of known minor vulnerabilities already reachable by other public models without special techniques. Amazon reported the method, coinciding with its dual role as investor and primary inference provider.

Anthropic's pattern of staged releases—Mythos Preview withheld, then Fable launched with guardrails—exposes the core tension in its safety posture. Pre-training scale gains in Fable produced measurable capability jumps in cybersecurity tasks, yet non-universal jailbreaks remain inevitable per the company's own blog. This mirrors prior alignment disputes where stated risk thresholds conflicted with deployment timelines.

The incident validates recurring government-Anthropic friction over successive model generations. Export controls now target weights and inference access rather than post-training only. Operational impact includes halted foreign employee workflows and customer re-routing to unaffected models while senior staff negotiate in Washington.

Next steps hinge on whether the specific bypass generalizes. Resolution within 14 days would restore access; extension past 30 days would force architecture changes to inference routing for compliance.