Anthropic releases Claude Fable 5 AI tool despite calling it 'too powerful'

Anthropic released Claude Fable 5 to the public this week—the same artificial intelligence system the company had previously insisted was too dangerous to distribute widely. The move marks a significant shift in strategy for the San Francisco AI firm, which had kept the model's more powerful sibling, Claude Mythos 5, restricted to a small circle of vetted partners since April.

Fable 5 represents Anthropic's most capable model yet available to general users. It operates in the same generation as Mythos 5, but with different safeguards and access restrictions. Both systems share a striking capability: they can work independently on human-assigned tasks for extended periods without direct oversight—a leap beyond anything the company has released before. When Anthropic announced the Fable launch on Tuesday, the company acknowledged the tension at its core: "launching a model with these capabilities carries risks," the statement read, even as it promised guardrails and limitations would be in place.

The decision to release Fable 5 comes against a backdrop of genuine concern about what these systems can do. When Mythos 5 was first handed to a small group of organizations for testing, Anthropic explained the restriction by pointing to the model's potential danger—its ability to find and exploit computer vulnerabilities, to breach systems. Canada's Finance Minister François-Philippe Champagne told the BBC in April that the scrutiny was warranted because Mythos represented "the unknown of the unknown." Yet even as Anthropic has refused to allow U.S. government agencies to use its AI tools—a dispute that has led to ongoing litigation with the Department of Defense—American government bodies have nonetheless been testing Mythos 5 themselves.

The early-access group testing Mythos 5 has already uncovered more than 10,000 critical security flaws in their systems. That discovery alone underscores why some observers worry the tool poses genuine risks to financial stability and cybersecurity infrastructure. Others, however, have suggested that much of the alarm surrounding these models may be partly marketing—a way to build mystique and investor confidence as Anthropic approaches a valuation near one trillion dollars and prepares for a public stock offering.

On Tuesday, Anthropic also announced that the roughly 150 organizations with early access to Mythos would now gain entry to Claude Mythos 5 without the previous restrictions on cybersecurity and biology applications—though access remains limited to what the company calls "a small group of cyber defenders and infrastructure providers" for now. The company indicated it plans to expand access through a broader "trusted access program" in the coming months. Each expansion of capability, each widening of the circle, strengthens Anthropic's position with investors betting on the future of AI.

But there is a growing unease in the room. Jack Clark, one of Anthropic's co-founders, appeared on BBC Newsnight last week to articulate a concern that has been building inside the company: the pace of AI advancement has become so rapid that the public ought to have some mechanism to slow it down. "It's important to have the option to take your foot off the accelerator and put it on the brake," Clark said. "Right now, it's as if the AI industry has an accelerator pedal but no brake pedal at all." The release of Fable 5—powerful, capable, and now in the hands of millions—suggests that brake may be harder to find than ever.