Anthropic Rolls Out Claude Mythos 5 AI Mannequin—Alongside With the Safer Fable 5 for the Public – Decrypt

Anthropic on Tuesday introduced the official launch of Claude Mythos 5, a restricted-access AI mannequin the corporate says has unprecedented cybersecurity capabilities, alongside Claude Fable 5, a model designed for broader public use.

In a weblog put up, Anthropic mentioned Mythos 5 will initially be accessible solely to accepted cybersecurity organizations, essential infrastructure operators, authorities companions, and chosen life sciences researchers. The corporate says the Mythos-class AI fashions “have reached a threshold the place they current important dangers.”

“In April, we started Undertaking Glasswing, releasing the primary Mythos-class mannequin (Claude Mythos Preview) to solely a restricted group of cyber defenders and demanding software program infrastructure suppliers,” Anthropic wrote. “After we did so, we said that we hoped to ultimately launch Mythos-level capabilities to all our customers, as long as we had developed new safeguards that have been robust sufficient to reliably forestall misuse.”

The launch follows months of concern over Mythos’ means to determine software program vulnerabilities and conduct complicated cybersecurity operations after information of the AI leaked on-line in March. Earlier this yr, the U.Ok.’s AI Safety Institute reported {that a} preview model turned the primary AI mannequin to finish a 32-step company community intrusion train with out human help. In April, Mozilla reported {that a} preview model of Mythos found over 271 vulnerabilities within the Firefox browser.

Till now, Anthropic has restricted entry to Mythos by its Undertaking Glasswing program whereas researchers evaluated the dangers. In a follow-up put up on X, Anthropic mentioned it might proceed to broaden entry to Mythos “by a broader trusted entry program,” for cybersecurity work and biomedical analysis.

The launch of Mythos additionally comes as tensions between Anthropic and President Donald Trump’s administration seem like easing after a public feud over the usage of Claude for navy operations.

“In session with the U.S. authorities, we plan to steadily broaden entry to Claude Mythos 5, persevering with our periodic addition of recent companions, in addition to pursuing a trusted entry program that enables cybersecurity organizations to use in a extra systematic method,” Anthropic wrote, including that customers of Mythos Preview can improve to Mythos 5.

Earlier this month, press stories mentioned that Anthropic had embedded engineers on the Nationwide Safety Company to assist deploy its Mythos AI mannequin for cybersecurity operations, reportedly focusing on China and Iran.

Alongside Mythos 5, Anthropic launched Claude Fable 5, a Mythos-class AI mannequin that makes use of the identical core design of Mythos 5 however with extra safeguards for most people.

“Fable 5’s capabilities exceed these of any mannequin we’ve ever made typically accessible,” Anthropic wrote. “It’s state-of-the-art on almost all examined benchmarks of AI functionality, displaying distinctive efficiency in software program engineering, information work, imaginative and prescient, scientific analysis, and plenty of different areas.”

In accordance with Anthropic, Mythos 5 and Fable 5 achieved high scores on a number of software program engineering, cybersecurity, biology, and reasoning benchmarks. The corporate reported a 78% rating on ExploitBench, which measures vulnerability discovery and exploitation, and an 88% rating on the Terminal-Bench coding benchmark. Anthropic mentioned the outcomes present the fashions can deal with longer and extra complicated duties than earlier Claude techniques.

Whereas Anthropic mentioned Fable 5 performs particularly properly on prolonged, complicated duties, with a view to forestall misuse, requests involving cybersecurity, biology, chemistry, or AI mannequin replication are dealt with by Claude Opus 4.8—a safeguard the corporate mentioned is triggered in fewer than 5% of classes.

“Releasing a mannequin this succesful comes with dangers,” Anthropic wrote. “With out safeguards, Fable 5’s capabilities in areas like cybersecurity could possibly be misused to trigger critical injury.”

The launch comes as AI builders face rising questions on methods to launch more and more succesful techniques. Anthropic has argued that Mythos may assist defenders discover and repair vulnerabilities earlier than attackers exploit them. On the identical time, critics, together with OpenAI CEO Sam Altman, have accused the corporate of utilizing “fear-based advertising and marketing” to promote its AI fashions.