An AI tool that its own creator said was too powerful to be released to the public is now available to anyone. Anthropic on Tuesday launched Claude Fable 5, a version of its Mythos class AI that had been restricted for months over cybersecurity fears. “Fable’s capabilities exceed those of any model we’ve ever made generally available,” the company said, adding that “releasing a model this capable comes with risks.”
When Mythos was first previewed privately in April, Anthropic warned that it was so intelligent it could exploit or hack computer systems. The warning triggered alarm among technology, finance and government leaders. Canadian finance minister François-Philippe Champagne told the BBC the attention was warranted because “it’s the unknown, unknown.” Despite an ongoing lawsuit between Anthropic and the US Department of Defense over government use of its AI, US agencies have been testing Mythos. The company’s valuation has neared $1tn.
“Anthropic releases Fable 5, a version of its powerful Mythos AI that was restricted for months over cybersecurity risks.”
Fable 5 is the first widely available model from the Mythos line, promoted for writing and debugging code, answering research questions and analysing images. But Anthropic built in strict safeguards: most queries about cybersecurity, biology or chemistry are diverted to the less capable Opus 4.8 model, which was released in late May. The company said it had also identified large-scale attempts to extract its technology to train competing AI in authoritarian countries, and those queries too will fall back to the weaker model.
Meanwhile, the roughly 150 organisations that had already accessed Mythos will now receive Claude Mythos 5, which removes limitations on cybersecurity and biology depending on the user’s needs. That group has expanded to about 200 organisations in more than 15 countries, including cybersecurity partners in Anthropic’s Project Glasswing program. Those partners reported finding more than 10,000 critical security flaws in their systems. Anthropic said it intends to expand access further through a “broader trusted access program”.
To test whether its restrictions could be bypassed, the company hired outside experts who spent more than 1,000 hours red‑teaming the model. Anthropic also runs a bug bounty programme. The US government, which had been in a legal dispute with Anthropic, has since tested Mythos. The White House set up an arrangement to test the most powerful models from leading AI companies before release, and Mythos 5 is being deployed in collaboration with the US government.
Anthropic co‑founder Jack Clark told the BBC last week that AI capabilities were expanding so rapidly that the public needed a way to slow the technology’s advance. “You want the option to be able to take your foot off the gas and put your foot on the brake,” he said. With Fable 5 now in millions of hands, that brake may already have been released.