Claude Fable 5 returned on July 1, 2026 — exactly 19 days after the U.S. government forcibly shut down Anthropic’s most powerful public AI model. The U.S. Department of Commerce lifted the export restrictions on June 30, and the very next day Fable 5 resumed operation globally across all Anthropic platforms. But the comeback happened with important changes: a new safety classifier, a different access model, and the parallel release of an entirely new model — Claude Sonnet 5.
Why Fable 5 Was Blocked and What Triggered It
To understand what changed, we need to recall the beginning of the story. On June 9, Anthropic released Claude Fable 5 — the first public Mythos-class model. We covered the model’s launch and blocking in detail in previous TechVisor articles.
The trigger was a discovery by Amazon researchers: a jailbreak method through a code-analysis request made it possible to bypass Fable 5’s protective filters. The model began identifying real software vulnerabilities and even demonstrating their exploitation. On June 12, the Department of Commerce issued a directive — and since Anthropic could not verify citizenship in real time, it disabled the model for everyone.
What Anthropic Did in 19 Days
During almost three weeks of downtime, Anthropic carried out extensive work on updating the model’s safety system and formalizing its relationship with the government.
New Safety Classifier
The main result is an updated automated classifier trained to block the specific jailbreak method discovered by Amazon. According to Anthropic, the new system blocks this technique in more than 99% of cases.
Testing was conducted through a partnership with Mozilla: researchers tested the model’s ability to create real exploits for Firefox 147 vulnerabilities. The result was a zero percent success rate in creating fully functional working exploits.
Partnership With the Government and Industry Standards
Anthropic formalized agreements to provide federal researchers with early access to frontier models before public release. Together with Amazon, Microsoft and Google, the company is also formalizing industry standards for assessing jailbreak risks for regulators.
What Changed in the Access Terms for Fable 5
This is the key practical point for everyone who uses Claude. The return conditions differ significantly from the initial release.
Weekly Limit Instead of Unlimited Access
At the first release, Fable 5 was available without restrictions within the subscription. Now the model is included in the subscription with a weekly usage limit.
Claude Code v2.1.190 contained a telling line: “You’ve used your Fable 5 usage for this week.” The old phrase “purchased separately from your plan” disappeared completely — the model is no longer sold separately.
Fable 5 vs Mythos 5: Different Return Conditions
- Claude Fable 5 — returns globally for everyone from July 1
- Claude Mythos 5 — only for approved U.S. organizations with critical infrastructure through Project Glasswing
Mythos 5 remains unavailable to the general audience.
Claude Sonnet 5 — a New Model Alongside the Comeback
Anthropic also released Claude Sonnet 5 — a new model for agentic tasks: complex multi-step workflows that require planning and executing a series of actions.
Price: $2.00 / $10.00 per million tokens (input/output) with introductory rates until August 31, 2026.
Documented cases: Rakuten — automated code review of production pull requests; Zapier — CRM updates and sending messages to enterprise customers.
What This Episode Changed for the AI Industry
The 19-day blocking of Fable 5 set a precedent with far-reaching consequences.
Researcher Tanishq Abraham formulated the key question: “Will the U.S. government require approval for every frontier model release?” If so, this fundamentally changes the pace of development for all AI labs.
The geography is also telling: India is Anthropic’s second-largest market after the U.S. The company has signed agreements with TCS (50,000 employees) and Infosys — and selected Indian government agencies already had access to Mythos through Project Glasswing.
In Brief: The Key Facts About Claude Fable 5’s Return
- Claude Fable 5 returned on July 1, 2026 — after 19 days
- The new classifier blocks the jailbreak in 99%+ of cases; Mozilla test: 0% success
- New terms: weekly limit in the subscription instead of unlimited access
- Mythos 5 — only approved U.S. organizations through Project Glasswing
- Bonus: Claude Sonnet 5 was released for agentic tasks ($2/$10 per million tokens)
- Precedent: Anthropic now gives the government early access to models
FAQ
Why was Claude Fable 5 blocked? Amazon discovered a jailbreak method through a code-analysis request — the model identified real vulnerabilities and demonstrated their exploitation. On June 12, the government applied export restrictions.
When did Claude Fable 5 return? July 1, 2026 — after 19 days. The Department of Commerce lifted the restrictions on June 30.
What changed in access to Fable 5? The model is now included in the subscription with a weekly limit instead of unlimited access. A new safety classifier has also been installed.
What is Claude Sonnet 5? A new model for agentic tasks. Price: $2/$10 per million tokens with introductory rates until August 31, 2026.
The article was prepared by the TechVisor team — practical IT media for people.
