Claude Fable 5 is back. The US Commerce Department lifted export controls on Anthropic’s frontier model on June 30, restoring global access as of July 1 — exactly 20 days after the June 12 ban that sparked an international regulatory crisis.
The reversal came after Anthropic demonstrated a finding that undermined the original justification: the same jailbreak technique that triggered the ban — reportedly discovered by Amazon researchers — worked identically on Opus 4.8, GPT-5.5, and Kimi K2.7. Every model tested could produce the same vulnerability identification and exploit code. Fable 5 had no unique offensive capability worth restricting.
Commerce Secretary Howard Lutnick personally wrote to Anthropic co-founder Tom Brown confirming the removal. Anthropic’s commitments in exchange: proactive jailbreak hunting, coordinated government review of future frontier model releases, 24/7 monitoring of malicious use, and a dedicated HackerOne bug bounty for Fable 5 jailbreaks. Most significantly, Anthropic agreed to give the US government earlier access to test future models before public release — the closest thing to a formal pre-approval process that currently exists.
The ban’s reversal comes with a 50% free usage window through July 7 for Pro, Max, Team, and select Enterprise users. After July 7, standard pricing applies — $10 per million input tokens, $50 per million output tokens. Cloud providers AWS, Google Cloud, and Microsoft Foundry are re-enabling access “as quickly as possible.”
Sam Altman called the government’s phased approval pattern “bad news” — and with reason. GPT-5.6 Sol, Terra, and Luna currently sit in exactly the same restricted-preview situation Fable 5 just escaped. Every major US AI lab now faces unpredictable timelines and business disruption from ad-hoc review processes, while Chinese competitors face no equivalent restrictions.
Mythos 5, however, remains limited to roughly 100 vetted US critical infrastructure operators under the Project Glasswing framework. No general access date has been announced.
The 20-day episode exposed a uncomfortable truth: Washington still has no binding process for frontier model review, only improvised responses. As the dust settles, one question remains — if every model tested could do what Fable 5 was banned for, what was the point?