Anthropic’s most advanced AI model Fable 5 is officially back online after one of the strangest interruptions the industry has seen so far.
Just three days after launching Claude Fable 5 in June, the company was forced to pull it offline worldwide following an emergency U.S. government order over cybersecurity concerns.
Now, after 18 days in limbo, Fable 5 has returned–albeit with a much tighter set of guardrails.
The whole saga felt pretty unprecedented.
Fable 5 debuted on June 9 and was immediately praised by developers as one of the best coding and reasoning models available.
But excitement didn’t last long.
Researchers reportedly discovered a jailbreak technique that could push the model into identifying software vulnerabilities and generating exploit code, prompting U.S. officials to step in.
The problem wasn’t just the vulnerability itself.
The government’s directive restricted access by foreign nationals, but Anthropic had no practical way to instantly verify the nationality of millions of users–or even all of its own employees.
Rather than selectively blocking people, the company simply shut the model down for everyone.
Now it’s back with a few notable changes.
Anthropic says it has added new cybersecurity safety classifiers designed to prevent the exploit-generation behaviour that sparked the shutdown.
The company has also agreed to work more closely with U.S. government agencies on testing future frontier models before they’re publicly released, while helping establish a broader industry framework for handling serious AI safety issues without forcing full product shutdowns.
For everyday users, access has mostly returned.
Subscribers on Claude Pro, Max, Team and Enterprise plans began receiving Fable 5 from July 1, although Anthropic says the model will initially count for up to 50% of users’ weekly usage limits before moving to a usage-credit system.
Developers can also access it through Anthropic’s API, with cloud platform support gradually returning across Amazon, Google and Microsoft services.
While Fable 5 is back, Anthropic’s more powerful cybersecurity-focused sibling, Mythos 5, remains much more tightly controlled, with access largely restricted to vetted organisations working in cyber defence.
The episode is already shaping up as one of AI’s biggest regulatory moments to date.
Instead of another debate about chatbots writing essays, it became a real-world test of what happens when governments decide an AI model is simply too capable to stay online.
