Anthropic’s most dangerous AI model just fell into the wrong hands

Anthropic's Mythos AI model, a almighty cybersecurity instrumentality that the institution said could beryllium unsafe successful the incorrect hands, has been accessed by a "small radical of unauthorized users," Bloomberg reports. An unnamed subordinate of the group, identified lone arsenic "a third-party contractor for Anthropic," told the work that members of a backstage online forum got into Mythos via a premix of tactics, utilizing the contractor's entree and "commonly utilized net sleuthing tools."

The Claude Mythos Preview is simply a caller general-purpose exemplary that's susceptible of identifying and exploiting vulnerabilities "in each large operating strategy and each large web browser …

Read the afloat communicative astatine The Verge.