Fable & Mythos are finally coming back. Anthropic says global access returns across Claude sites today. Try it while it's free! Afte...
 |
| Fable & Mythos are finally coming back. Anthropic says global access returns across Claude sites today. Try it while it's free! |
Commerce Secretary Howard Lutnick's letter to Anthropic noted that the company had worked with the United States government to address risks and that the controls were withdrawn for now. The government reserves the right to reassess and reimpose restrictions if circumstances change.
The rollercoaster began on June 12, when the United States Department of Commerce imposed emergency export controls on both models, citing national security risks. The move was triggered by an Amazon research report that identified a jailbreak technique allowing Fable 5 to identify software vulnerabilities and generate exploit code. Unable to verify user nationality in real time, Anthropic was forced to suspend all access globally. The intervention marked the first known instance of export control authorities being used to pull AI software, rather than chips or hardware, from public access. For 18 days, developers and enterprises that had integrated Fable 5 were left scrambling.
New Safeguards
The restoration comes with a critical new layer of protection: an improved safety classifier that blocks the specific jailbreak technique in over 99 percent of cases. When a request is flagged—particularly in three protected areas: cybersecurity, biology and chemistry, and model distillation—the query is automatically routed to Claude Opus 4.8 instead of being answered by Fable 5.
Fallback mechanism. This is a deliberate design choice. Rather than issuing a hard refusal that would dead‑end legitimate research, the system quietly redirects to a model whose behavior in these areas is well understood and considered safe. The classifiers trigger in fewer than 5 percent of sessions on average, though Anthropic acknowledges they are tuned conservatively, meaning some harmless cybersecurity questions may also be redirected. For builders integrating Fable 5 via application programming interface, the fallback returns a normal HTTP 200 with stop_reason: "refusal". The company provides three ways to handle this: a server-side fallback parameter, SDK middleware, or manual retry.
Fable 5 versus Mythos 5
The return of the two models is not identical. Fable 5, designed for the mass market with robust safety mechanisms, is now available worldwide through Claude.ai, Claude Code, and Claude Cowork. To incentivize re‑adoption, Anthropic is offering a temporary rollout plan through July 7, with Fable 5 usage included at no added cost for up to 50 percent of a user's weekly tier allowance.
Mythos 5, the same underlying model but without the safety classifiers, remains a different story. While the export controls have been lifted, access is still restricted to vetted United States organizations through Project Glasswing, Anthropic's controlled‑access cybersecurity program. International partners previously part of the program, including Samsung Electronics, SK Hynix, and European banks, face delays in regaining access. Anthropic continues negotiating with the government to expand domestic and international access.
The episode is reshaping how frontier AI models are regulated. Anthropic has committed to expanded pre‑release government access for model evaluation, rapid disclosure of jailbreaks, dedicated joint research resources, and participation in a voluntary industry safety framework. The company is also working with Amazon, Microsoft, and Google to draft a consensus framework for assessing jailbreak severity. For now, one of the most powerful AI models ever released is back online, with new guardrails designed to keep it safe without stifling its remarkable capabilities.
