In a rapid reversal that has sent ripples through the tech sector, the U.S. government has officially lifted the emergency export controls imposed on Anthropic’s flagship artificial intelligence models. Effective immediately, global access is being restored to Claude Fable 5 and its restricted sibling, Claude Mythos 5, concluding a tense, two-week blackout that left developers and enterprise clients in limbo.
The decision represents a major sigh of relief for Anthropic and a significant de-escalation in the friction between the AI startup and federal regulators.
The Two-Week Global Blackout
The friction began when the U.S. Commerce Department issued an abrupt export control directive. The order required Anthropic to immediately cut off access to its newly released Fable 5 and Mythos 5 models for any "foreign national"—a designation that applied globally, inside or outside the U.S., and even included Anthropic’s own non-citizen staff.
Because the government's mandate took effect instantly, Anthropic faced an impossible technical hurdle: it had no reliable mechanism to verify the nationality of every user connecting to its platform in real time. Erring on the side of compliance, the company made the drastic choice to take both state-of-the-art models completely offline for all users worldwide.
The Trigger: An Amazon 'Jailbreak'
The emergency intervention was triggered by a vulnerability report from researchers at Amazon—a key Anthropic partner. The researchers discovered a "jailbreak" prompt that successfully bypassed Fable 5’s safety protocols. When executed, the exploit allowed the AI to identify specific software vulnerabilities and, in one instance, generate actionable code demonstrating how those security flaws could be actively abused.
While Anthropic downplayed the incident—noting that the same prompt routinely works on standard rival models like OpenAI’s GPT-5.5 and China's Kimi K2.7—federal authorities viewed the risk to national security as severe enough to justify pulling the plug.
You might also like: TKROBOTS AI: An AI-Powered Crypto Trading Platform Marketed as a Passive Income Tool
Engineering a Resolution
To break the regulatory gridlock, Anthropic spent the last two weeks collaborating intensely with federal agencies to build robust countermeasures. The company engineered an advanced safety filter—a specialized classifier—trained explicitly to recognize and neutralize the exact jailbreak methodology highlighted in the Amazon report.
According to Anthropic's technical brief, the new classifier successfully blocks the exploit in over 99% of attempts. When a potentially hazardous coding request triggers the filter, Fable 5 will automatically refuse the prompt and hand the session off to the older, weaker Claude Opus 4.8 model.
Impressed by the rapid fix, Commerce Secretary Howard Lutnick confirmed that Anthropic had successfully mitigated the underlying risks, officially rendering the need for an emergency export license obsolete.
Looking Ahead
With federal clearance secured, Claude Fable 5 is returning to the wild across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. For enterprise tiers and Pro users, Anthropic is initially including Fable 5 for a portion of their weekly limits before transitioning it to standard usage credits. The more tightly bound Mythos 5 model remains on a shorter leash, currently restricted to a vetted list of roughly 100 trusted U.S. corporations and government agencies.
Beyond getting its tech back online, the saga has pushed Anthropic, Amazon, Google, and Microsoft to begin co-developing a unified industry framework to uniformly judge, triage, and report jailbreak severity. As frontier AI continues to advance rapidly, this brief shutdown highlights a new era of proactive federal oversight—proving that tomorrow's AI dominance depends entirely on mastering today's safety filters.












































