Who leaked Mythos? Everything to know about Discord group behind Anthropic's AI breach

Anthropic released its highly advanced and most powerful model, “Claude Mythos,” in April 2026.

However, due to its exceptional capabilities in cybersecurity, it was not accessible for general public aiming to prevent misuse.

But recently, an unauthorized group of users in a private Discord chat gained access to the secret model. As reported by Bloomberg, the access was made possible due to a third-party contractor for Anthropic who used previously leaked knowledge to guess where the model was stored.

Since then, the group has been actively using the program. However, they have not made it accessible to the general public.

The company confirms that the anonymous users are not linked to any cyberattacks, confirming it is “investigating a report claiming unauthorized access” through a third-party vendor environment.

This loophole has made security analysts quite concerned. According to Mozilla's head technology officer, Bobby Holley, Mythos turns an intelligent engineer into a “world-class, elite security engineer.”

It was used by the company to discover and fix 271 different bugs that existed in the browser called Firefox. Anthropic also states that Mythos discovered a security flaw that is 27 years old in OpenBSD.