Anthropic restored Fable 5 worldwide on July 1 with its abilities intact, adding a filter that blocks a flagged hacking prompt in more than 99% of attempts.
Key Points:
- The U.S. Commerce Department lifted its export controls, and Fable 5 returned globally across Anthropic's main products.
- The model keeps its full capabilities, though a new filter reroutes one flagged hacking prompt to a weaker model.
- Testing showed cheaper models could find the same flaws, which weakened the case for treating Fable 5 as uniquely dangerous.
Fable 5 Returns Globally
The U.S. Commerce Department lifted the emergency export order on June 30, and the model reached users again the next day across the Claude Platform, Claude.ai, Claude Code and Claude Cowork. Access through the major cloud platforms will follow.
Paid subscribers can use it inside their weekly limits through July 7, after which continued access shifts to a metered usage-credit system.
The model had gone dark on June 12. Regulators barred every foreign national from touching it and gave the company only about 90 minutes to comply, so Anthropic pulled the model for all users rather than attempt to screen them one by one in real time.
The directive followed a report in which Amazon researchers found a prompt that steered the model into flagging a cluster of software vulnerabilities across common systems. In one instance it went further and produced working code that showed how a single flaw could be exploited. That single demonstration, more than the vulnerability list, is what set off the alarm.
Also Read: CrowdStrike Warns Claude Mythos Could Speed Zero-Day Decisions
Testing Cleared Fable 5
Anthropic's own review, run alongside the government and the firm that filed the report, confirmed that far weaker models could surface the same flaws, among them Opus 4.8, GPT-5.5 and Kimi K2.7. Every other model the team checked could still reproduce the lone exploit demonstration, which drained the case that Fable 5 alone held a singular cyber weapon. The company had called the ban an overreach.
The fix was narrow. On its return the model carries one classifier that catches the reported prompt and reroutes it to the weaker Opus 4.8, leaving Fable 5's abilities intact but raising false alarms on ordinary coding, and warning the user whenever it intervenes.
Mythos 5 Stays Limited
Mythos 5, the same underlying model with fewer guardrails, remains fenced off to roughly 100 vetted U.S. organizations inside a program called Project Glasswing.
The company also opened a public channel for jailbreak reports. It also promised regulators earlier access to test future frontier models before launch.
Fable 5 first launched on June 9 as Anthropic's first widely available Mythos-class model, praised across a range of industry benchmarks before it disappeared within days of release. The outage lasted barely three weeks. Its comeback now reclaims the benchmark ground that a fast-rising Chinese lab had quietly taken over while the model sat idle.
Read Next: XRP Tests $1.07 Again As June Rejections Shadow Bulls