AI Once Deemed Too Dangerous Now Released to Public

Anthropic has released Claude Fable 5, a public version of an AI system that previously raised concerns among technology, finance and government leaders because of its advanced capabilities.

The model is based on Claude Mythos, an AI tool the company first released privately in April to a limited group of organizations for testing and preview. At the time, Anthropic said the system was too powerful for broad public release because of its ability to identify and potentially exploit weaknesses in computer systems.

On Tuesday, Anthropic said Fable 5 would now be made available to the public, but with safeguards and user limits in place. The company acknowledged the risk, saying that releasing a model this capable carries serious concerns.

According to Anthropic, Fable’s capabilities are stronger than any model the company has previously made generally available. The public version is designed with restrictions, while Mythos 5, a less limited version of the same underlying model, will remain available only to selected organizations.

The original private release of Mythos drew attention because of its cyber capabilities. Some officials and experts warned that such tools could create financial and security risks if misused, while others questioned whether the concern reflected genuine danger or part of the broader marketing competition surrounding advanced AI.

Canadian Finance Minister François-Philippe Champagne said in April that the concern around Mythos was justified because of the uncertainty surrounding such powerful systems.

Despite Anthropic’s legal fight with the U.S. Department of Defense over government use of its AI tools, U.S. agencies have also reportedly tested Mythos. The company is expected to go public soon, with its private valuation approaching $1 trillion, making each major AI release important for investors watching the company’s position in the race.

Anthropic said around 150 groups that had early access to Mythos will now receive access to Claude Mythos 5. Depending on the organization and use case, that version may not include the same restrictions on cybersecurity or biology-related work.

The company said users of Mythos have already reported finding more than 10,000 critical security flaws in their own systems. Anthropic said the more powerful access is currently limited to a small group of cyber defenders and infrastructure providers, though it plans to expand access through a broader trusted program.

Both Fable and Mythos can work on human instructions for longer unattended periods than previous Claude models, according to Anthropic.

Anthropic co-founder Jack Clark recently warned that AI development is accelerating so quickly that society may need a way to slow it down. He said the industry currently has a gas pedal, but not a brake pedal.