A version of an artificial intelligence tool that its own creator said was too powerful to release to the public has just been released to the public. Claude Fable 5, a version of Anthropic's Claude Mythos, is now available after the company previously warned the AI was so intelligent it could exploit or hack computer systems.
Anthropic said on Tuesday that Fable will be released with safeguards and user limitations in place, but acknowledged in a statement: "Releasing a model this capable comes with risks." The company added that "Fable's capabilities exceed those of any model we've ever made generally available."
“Anthropic releases Claude Fable 5, a version of its AI tool it previously deemed too dangerous for public use.”
The tool caused serious concerns among technology, finance, and government leaders when it was first released privately in April for previewing and testing. Canadian Finance Minister François-Philippe Champagne told the BBC in April that the attention on Mythos was warranted, in part, because "it's the unknown, unknown."
Despite Anthropic's ongoing lawsuit against the US Department of Defense over its refusal to accept government use of its AI tools, US government agencies have been testing Mythos. The company is expected to become a public company soon, as its private valuation has neared $1tn (£747bn).
On Tuesday, Anthropic also said that the roughly 150 groups that had been given access to preview Mythos will now have access to Claude Mythos 5, which does not have limitations on cybersecurity or biology, depending on an organization's specific uses. To date, groups and companies using Mythos have reported finding more than 10,000 critical security flaws in their systems.
Anthropic said the updated access was limited to a "small group of cyberdefenders and infrastructure providers" but is expected to go beyond that soon. "We intend to expand access to Mythos 5 through a broader trusted access program," the company said.
Both Fable and Mythos, which are essentially the same model but with different safeguards and levels of access in place, can work "unattended" on human commands for longer periods than any previous Claude models, Anthropic added.
Anthropic co-founder Jack Clark last week told BBC Newsnight that the ability of AI tools was expanding so rapidly that the company thought there should be a way for the public 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.