Anthropic Releases Claude Fable 5 AI Tool Despite Prior Security Concerns

Anthropic has officially released Claude Fable 5, an advanced artificial intelligence tool, to the public. This decision comes despite the company’s earlier assertions that a version of this AI, Claude Mythos, was too potent for general release, citing significant security risks.

Claude Mythos, the precursor to Fable 5, garnered considerable attention among technology, finance, and government circles upon its private release in April for preview and testing. Concerns were voiced by senior figures that its capabilities could pose substantial financial security threats, although some critics dismissed such warnings as marketing hyperbole.

Anthropic stated on Tuesday that Fable 5 would include safeguards and user limitations, yet conceded, “releasing a model this capable comes with risks.” The company explicitly noted that “Fable’s capabilities exceed those of any model we’ve ever made generally available.” Initially, Mythos was confined to a select group of organisations due to its purported intelligence, including its capacity to exploit or hack computer systems.

Canadian Finance Minister François-Philippe Champagne remarked in April that the focus on Mythos was justified, highlighting the “unknown, unknown” aspects of its power. Concurrently, US government agencies have also been testing Mythos, notwithstanding Anthropic’s ongoing legal dispute with the US Department of Defense over its refusal to sanction governmental use of its AI tools.

Anthropic, with a private valuation approaching $1 trillion, is anticipated to go public soon. Demonstrating enhanced and broader AI capabilities serves to underscore its value to investors. The company also announced that the approximately 150 groups previously granted preview access to Mythos would now receive access to Claude Mythos 5, which lacks the cybersecurity or biology limitations imposed on the public version, depending on specific organisational uses. To date, these groups have reported identifying over 10,000 critical security flaws within their systems using Mythos.

Access to Mythos 5 is currently restricted to a “small group of cyberdefenders and infrastructure providers,” with plans to broaden this through a “trusted access program.” Both Fable and Mythos, essentially the same model with varying safeguards and access levels, can operate “unattended” on human commands for extended periods, a capability exceeding prior Claude models.

Anthropic co-founder Jack Clark last week suggested the rapid expansion of AI capabilities necessitated a public mechanism to decelerate technological advancement. Clark posited, “You want the option to be able to take your foot off the gas and put your foot on the brake. Right now, it’s like the AI industry has a gas pedal, but it doesn’t have a brake pedal.”