Anthropic has restored global access to Claude Fable 5 after the U.S. Department of Commerce lifted export controls, ending an 18-day standoff triggered by a jailbreak technique flagged by Amazon researchers. The model is now live on claude.ai, Claude Code, and Claude Cowork, with API access on AWS, Google Cloud, and Microsoft Foundry expected soon, while paid users receive up to 50% of their weekly usage limits dedicated to the model through July 7.
Anthropic restores Claude Fable 5 globally after 18-day export control standoff
Anthropic brought Claude Fable 5 back online today. The move comes less than 24 hours after the U.S. Department of Commerce officially lifted export controls on the model, ending a chaotic two-week standoff that forced the AI lab to pull its flagship system worldwide.
It started on June 12, when the Commerce Department issued an emergency directive barring any foreign national from using Fable 5 or its parent architecture, Mythos 5. The rule was absolute. It even covered Anthropic's own non-citizen employees. Faced with the impossible task of verifying every user's nationality in real time, Anthropic simply shut both models down.
The trigger for the shutdown was a specific technique flagged by Amazon researchers. They demonstrated that Fable 5 could be prompted to identify software vulnerabilities and write exploit code. Not exactly a feature for a general-purpose chatbot.
The fix and the fallout
The resolution involved a single safety filter tuned to block that exact technique. Anthropic trained a new classifier that catches the reported prompt in over 99% of cases. If the filter trips, you get a notification and your request gets rerouted to the older Opus 4.8.
Keep in mind that this approach is inherently fragile. It targets one known prompt, not the model's underlying capability. Fable 5 can still find those vulnerabilities; it just can't show you how to build one using the specific method Amazon reported. Detection-based safeguards like this were the whole reason the government got nervous in the first place, and a classifier tuned to a single bypass does nothing for the next one.
Anthropic's own review found that GPT-5.5, Kimi K2.7, and even Opus 4.8 could reproduce the exploit demonstration. Multiple industry voices agreed, with Corridor CPO Alex Stamos publicly stating that kneecapping one of America's AI champions during a race with Beijing was "incredibly stupid."
What's available now
Fable 5 is back on claude.ai, Claude Code, and Claude Cowork starting today. If you're on a Pro, Max, Team, or select Enterprise plan, you'll get up to 50% of your weekly usage limits dedicated to Fable 5 through July 7. After that, access shifts to usage credits.
API access on AWS, Google Cloud, and Microsoft Foundry will follow "as quickly as possible," though Anthropic didn't pin down a specific date. The timeline is still blurry. Legacy cloud integrations of this scale rarely move at consumer app speed, so if you're building enterprise pipelines on those platforms, you might want to monitor the release notes rather than holding your breath for a weekend update. The more restricted Mythos 5, which shares Fable 5's architecture but strips back safety guardrails, was already restored on June 26 for vetted U.S. organizations under Project Glasswing.
Anthropic isn't promising perfection. The company concedes that making any AI model fully immune to jailbreaks is probably impossible. To that end, they've launched a dedicated HackerOne program for security researchers to submit potential Fable 5 bypasses and committed to pre-releasing frontier models for government testing.
