TL;DR — On July 1, 2026, Anthropic redeployed Claude Fable 5 globally after the US government lifted export controls imposed on June 12. The company simultaneously published a detailed breakdown of its cyber safety classifiers and proposed an industry-wide jailbreak severity framework — co-developed with Amazon, Microsoft, Google, and other Glasswing partners — alongside a new HackerOne bug bounty program for jailbreak reporting.
What happened
On June 12, the US government applied export controls to Anthropic’s newest models — Claude Fable 5 and Claude Mythos 5 — requiring the company to restrict access to foreign nationals. Because the order took effect immediately and Anthropic had no reliable real-time nationality verification, it suspended access to both models for all users. The freeze lasted nearly three weeks.
The trigger: Amazon researchers found a technique to bypass Fable 5’s safety classifiers, prompting it to identify software vulnerabilities and — in one case — produce demonstration exploit code. The government responded with export restrictions.
Anthropic’s own testing, however, revealed that the reported technique did not expose unique capabilities. Every model tested — including Claude Opus 4.8, GPT-5.4, GPT-5.5, and Kimi K2.7 — could produce the same outputs.
On June 30, the export controls were lifted. Fable 5 went live globally on July 1 across Claude.ai, Claude Platform, Claude Code, and Claude Cowork. Mythos 5 was restored for a set of US organizations approved under the Glasswing program.
Stronger safeguards
Anthropic detailed its “defense in depth” approach: classifiers trained to block four categories of activity — prohibited uses (ransomware, malware development, data exfiltration), high-risk dual uses (exploit development, penetration testing), low-risk dual uses (vulnerability scanning), and benign uses. Fable 5’s safety margin was deliberately set wider than any previous model, blocking more benign requests to catch more harmful ones.
A new classifier, trained in collaboration with the government, now blocks the Amazon-reported jailbreak technique in >99% of cases. NIST’s CAISI independently validated the safeguards.
A jailbreak severity framework
The most forward-looking element is Anthropic’s draft framework for scoring jailbreak severity. There is currently no industry standard for rating how dangerous a given jailbreak is — some merely unblock minor behaviors, while universal jailbreaks dismantle a model’s entire safety envelope.
The proposed framework categorizes jailbreaks on a spectrum: minor intrusions into the safety margin, narrow harmful jailbreaks (unblocking specific dangerous behaviors), and universal jailbreaks (unblocking a wide range of harms). Anthropic published the draft publicly and opened a HackerOne program inviting security researchers to submit jailbreaks against Fable 5.
Why it matters
This episode illustrates a recurring tension: frontier AI models are being regulated like munitions, with escalating safeguards and government oversight. Anthropic’s response — publishing the classifier taxonomy, proposing an industry standard, and opening a public bug bounty — treats safety not as a competitive secret but as shared infrastructure.
Whether the jailbreak framework gains traction remains to be seen. But with Glasswing partners Amazon, Microsoft, and Google already at the table, Anthropic is betting that the industry is ready to standardize how it talks about — and fixes — AI safety failures.
FAQ
Q: Is Fable 5 available to everyone now? Yes. Fable 5 is available globally on Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. Pro, Max, Team, and select Enterprise users get up to 50% of weekly usage limits included through July 7, after which Fable 5 access requires usage credits.
Q: What about Mythos 5? Mythos 5 is only available to US organizations approved under Project Glasswing. Broader access is pending government coordination.
Q: Can I report a jailbreak I find?
Yes. Anthropic launched a HackerOne program at hackerone.com/anthropic. Security researchers can submit jailbreaks against Fable 5 for review.
Q: What happened to Claude Opus 4.1? Separately, Anthropic deprecated Claude Opus 4.1 (claude-opus-4-1-20250805) with retirement on the API scheduled for August 5, 2026.