Anthropic's Claude Fable 5 and Mythos 5 are back after 18 days offline, following a Commerce Department review and new safety fix.
Anthropic's Claude Fable 5 and Mythos 5 models are back online after the U.S. government lifted export controls that had blocked access to both worldwide for 18 days, the company said Tuesday. Starting Wednesday, Fable 5 returns across Claude.ai, Claude Code, Claude Cowork and the Claude platform, closing out one of the more disruptive stretches in the company's recent history.
At a Glance
- Fable 5 access resumes Wednesday on Claude.ai, Claude Code, Claude Cowork and the Claude platform
- Pro, Max, Team and select Enterprise subscribers get Fable 5 usage up to half their weekly allotment through July 7
- Mythos 5 access restored for a limited group of U.S. organizations following government approval on June 26
- New safety classifier blocks the flagged exploit technique in over 99 percent of tested attempts
- Cloud reinstatement on AWS, Google Cloud and Microsoft Foundry is planned but has no confirmed date
Why the Models Went Dark
The trouble traces back to June 12, when the Commerce Department issued a directive after Amazon researchers documented a method for coaxing Fable 5 into producing dangerous outputs tied to software vulnerability discovery. The order applied broadly to foreign nationals, a definition wide enough to sweep in researchers working inside Anthropic itself. With no practical way to comply while keeping the models running, Anthropic pulled Fable 5 and Mythos 5 offline entirely rather than risk a narrower, unworkable fix.
Anthropic pushed back on the idea that its models were uniquely at fault. The company said its internal testing showed that other mainstream systems, including OpenAI's GPT-5.5, could be prompted into generating the same flagged outputs described in Amazon's report. That detail matters because it frames the shutdown less as a Fable specific flaw and more as an industry wide vulnerability that happened to surface first in Anthropic's product.
The Fix and the Sign Off
In response, Anthropic built a new safety classifier designed specifically to catch the exploit technique Amazon had documented. In testing, the company said the classifier neutralized the bypass in more than 99 percent of attempts. The Commerce Department's Center for AI Standards and Innovation reviewed both the original safeguards and the new classifier, and Anthropic said the results confirmed the fix works.
That review appears to have satisfied officials at the top of the government. Commerce Secretary Howard Lutnick said on X that his department had spent two weeks working alongside Anthropic to vet Fable 5 and align the review across federal agencies, with the stated aim of strengthening American standing in the technology race. White House chief of staff Susie Wiles echoed that framing, telling X, according to The Wall Street Journal, that the shared goal was deploying capable technology
