Anthropic released Claude Fable 5, the most capable model it has ever made generally available, on 9 June 2026. Three days later, on 12 June, a US government export control directive forced the company to disable both Fable 5 and its restricted sibling, Mythos 5, for every customer worldwide. The models Anthropic had spent the week describing as state-of-the-art were offline within 72 hours of launch.
What Anthropic launched
Fable 5 and Mythos 5 are the same underlying model, separated only by their safeguards. Fable 5 was the public version, sold through the Claude API and bundled into Pro, Max, Team and Enterprise subscriptions during an introductory window. Mythos 5 — the same model with certain safeguards removed — was limited to a small group of vetted cybersecurity partners under a programme Anthropic calls Project Glasswing, with biomedical researchers due to follow. Access to either cost US$10 per million input tokens and US$50 per million output tokens, roughly double the price of Claude Opus 4.8. Anthropic said Fable 5 was “state-of-the-art on nearly all tested benchmarks of AI capability”, citing software engineering, vision and scientific research; those claims have not been independently verified.
How the safety limits worked
Rather than build its guardrails into the model itself, Anthropic ran Fable 5 alongside separate classifier systems. When a request fell into one of three areas, the system handed off to the older Claude Opus 4.8:
- Offensive cybersecurity, including exploitation and agentic hacking
- Bioweapons queries and dual-use biology or chemistry
- Attempts to distil the model's capabilities into competing models
Anthropic said more than 95% of Fable sessions involved no fallback at all, and that external bug-bounty testing found no universal jailbreaks across more than 1,000 hours. A mandatory 30-day data retention policy applied to all Fable 5 and Mythos 5 traffic.
The export order
The directive arrived on Friday 12 June at 5:21pm ET. According to TechCrunch, the US Commerce Department invoked export control powers to bar access by any foreign national, whether inside or outside the United States, including Anthropic's own foreign-national employees. To ensure compliance, Anthropic disabled both models for all customers rather than attempt to screen by nationality, and said its other models, including Claude Opus 4.8, were unaffected. Search Engine Journal reported that some Claude Max subscribers who had paid specifically for Fable 5 access asked for refunds.
Anthropic's objection
Anthropic said it believed the order stemmed from a “potential narrow, non-universal jailbreak” — a method of bypassing the cybersecurity guardrail by asking the model to review a codebase for security flaws rather than to fix them. The company said it received only verbal evidence, and the government's letter has not been made public. Anthropic disputed that the finding justified pulling the model, saying it disagreed “that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people”, and arguing that the same capability already exists in rival public models and is used routinely by defenders. Katie Moussouris, a security researcher shown the underlying paper, told TechCrunch the bypass should never have triggered an export control.
What happens next
The Commerce Department has not published its reasoning, and CNBC reported that Anthropic was set to meet the Trump administration to resolve the dispute. Cybersecurity veterans publicly criticised the ban, arguing that removing a defensive tool weakens the people who use it to find software flaws before attackers do. Some reporting has linked the action to friction between the company and the administration rather than the jailbreak alone, though the government has not confirmed any rationale. For now the episode sets an unusual precedent: a frontier model withdrawn from the public not by its maker but by government order, days after release. Whether Fable 5 returns, and on what terms, depends on a rationale that neither side has yet disclosed.
Sources
- Anthropic — Claude Fable 5 and Claude Mythos 5
- TechCrunch — Anthropic releases Claude Fable 5 to the public (9 June 2026)
- TechCrunch — The government has pulled the plug on Anthropic's most powerful AI (12 June 2026)
- TechCrunch — The Anthropic models ban was never about an AI jailbreak (15 June 2026)
- Search Engine Journal — Government order shuts down Fable 5 despite Anthropic's objections (13 June 2026)
- CNBC — Anthropic to meet with Trump administration over Mythos dispute (15 June 2026)
