The US government just lifted the most consequential AI export controls in history.
Anthropic announced on X that the Department of Commerce has lifted export controls on Claude Fable 5 and Claude Mythos 5. Access restoration begins tomorrow, and the company says it will share a full update soon.
This is the full reversal of the June 12 export controls that took both models offline globally. At the time, the government directed Anthropic to suspend all access to Fable 5, its most capable general model, and Mythos 5, its flagship cybersecurity model, citing national security concerns under a new AI export control framework.
The Two-Week Saga in Brief
The story broke on June 12 when the US government killed access to Anthropic’s best models in what was the first use of software-specific export controls against an AI company. Within days, reporting revealed that Amazon CEO Andy Jassy had personally raised concerns about Anthropic’s models to Trump administration officials, conversations that triggered the entire crackdown.
The situation evolved rapidly. On June 18, I covered why security experts said the government’s demand for jailbreak-proof AI was impossible to meet. Then on June 26, Anthropic got a partial win: Mythos 5 could be redeployed to US critical infrastructure organizations, but Fable 5 remained locked.
Now both are free, and Anthropic has published the full story of what actually happened.
How Amazon Researchers Triggered the Export Controls
Anthropic published the complete account of what triggered the June 12 directive. The government became aware of a report in which Amazon researchers found a method of bypassing Fable 5’s safeguards. The researchers prompted the model to identify software vulnerabilities, and in one case it produced code demonstrating how to exploit one of them.
The critical detail: Anthropic says every model they tested could reproduce the same behavior. Claude Opus 4.8, GPT-5.5, Kimi K2.7 — all of them could identify those vulnerabilities. Every single model they tested, including Haiku 4.5, Sonnet 4.6, and Opus 4.6 through 4.8, could produce the same exploit demonstration. The technique did not expose any Mythos-level offensive capabilities.
New Classifiers and the Opus 4.8 Fallback
Anthropic worked with the government to train an improved safety classifier targeting the specific behavior from the Amazon report. It blocks the technique in over 99% of cases. Researchers from the US Department of Commerce’s Center for AI Standards and Innovation (CAISI) have tested both the old and new safeguards and say they are extraordinarily strong.
When a request to Fable 5 gets blocked by the classifier, instead of just rejecting it, the system silently falls back to Opus 4.8. The user gets their answer from a less capable model rather than nothing. Anthropic says it will continue refining the classifiers to reduce false positives during routine coding and debugging.
An Industry Framework for Jailbreak Severity
Anthropic announced it is partnering with Amazon, Microsoft, Google, and other Glasswing partners to draft a consensus framework for assessing how serious an AI jailbreak actually is. The proposal scores jailbreaks on four criteria: how much extra capability the jailbreak provides, how broad the capability gain is, how easy it is to weaponize, and how easy it is to discover.
The company also launched a HackerOne program where security researchers can submit potential Fable 5 jailbreaks for review once the model is available again.
Deeper Government Collaboration
Anthropic committed to a significant expansion of its government partnership. The four-part agreement includes pre-release access to frontier models for independent government evaluation, rapid information sharing when jailbreaks are discovered, dedicated compute and expertise for joint research, and a push for a common voluntary security standard across frontier model providers.
Anthropic said these rules should be codified in strong regulation and applied equally across all frontier AI developers.
What This Means
Fable 5 will be available starting tomorrow, July 1, to users globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. For Pro, Max, and Team plans, it counts toward up to 50% of weekly usage limits through July 7, after which it will use usage credits.
Mythos 5 was already restored for US organizations on June 26. Anthropic continues working to expand access through the Glasswing program.
For the broader AI industry, this saga sets an important precedent. The US government imposed controls, got engagement from Anthropic on safety measures, and reversed them in under three weeks. The outcome is not just a restored model but the beginnings of an industry-wide framework for handling jailbreak severity. The open question is whether this collaboration model scales to every frontier model developer, or only the ones with existing government relationships.
Bottom Line
The export controls on Claude Fable 5 and Mythos 5 are gone. Anthropic published the full story of how an Amazon researcher’s jailbreak report triggered the entire crisis, built new classifiers to address it, and is now working with Amazon, Microsoft, and Google on a shared framework for evaluating jailbreak severity. A rapid, dramatic reversal that ended up producing more transparency and industry coordination than the crisis itself might have otherwise generated.
