A version of an artificial intelligence tool that the company said was too powerful to be released to the public has just been released to the public. Claude Fable 5, a "safe" version of Anthropic's Claude Mythos AI program, became available on Tuesday, months after Mythos was restricted to a small set of partner institutions over cybersecurity fears.
The Mythos model caused serious concern among technology, finance and government leaders when it was privately released for previewing and testing in April. "Releasing a model this capable comes with risks," Anthropic said on Tuesday. "Fable's capabilities exceed those of any model we've ever made generally available."
“Anthropic released "safe" Fable 5 AI to public after months of restricting its powerful Mythos model over cybersecurity fears.”
When Mythos was first unveiled, Anthropic said it was so intelligent it could be dangerous because of its ability to exploit or hack computer systems. Canadian finance minister François-Philippe Champagne told the BBC in April that the attention was warranted, in part, because "it's the unknown, unknown."
Despite a lawsuit between Anthropic and the US Department of Defense over government use of its AI tools, US agencies have been testing Mythos. The White House has set up an arrangement to test the most powerful models from leading AI companies before release. Mythos 5 is being deployed in collaboration with the US government.
To-date, groups and companies using Mythos have reported finding more than 10,000 critical security flaws in their systems. The roughly 150 organisations that previously had access to Mythos will now receive Claude Mythos 5, which does not have limitations on cybersecurity or biology, depending on their specific uses.
For the public version, Fable 5 includes safeguards: most queries about cybersecurity, biology or chemistry are routed instead to a lower-tier model called Opus 4.8. Anthropic said it had identified large-scale attempts to extract its technology to train competing AI models in authoritarian countries; those queries will also fall back to the less capable model. The company hired outside experts to spend more than 1,000 hours trying to bypass these restrictions, a process known as red-teaming.
Anthropic is expected to become a public company soon, with its private valuation nearing $1tn (£747bn). Both Fable and Mythos can work "unattended" on human commands for longer periods than any previous Claude models.
Anthropic co-founder Jack Clark last week told BBC Newsnight that the expansion of AI tools was so rapid that the public should have a way to slow the technology's advancement. "You want the option to be able to take your foot off the gas and put your foot on the brake," he said.