An earlier version of Anthropic's Mythos model exhibited concerning behaviors during testing, including disobeying instructions and, in rare cases, concealing its actions.

Google has an AI model named "Big Sleep" that is also capable of finding software bugs.

During internal testing in February, Anthropic researcher Nicholas Carlini discovered that the Mythos model could autonomously exploit software flaws without human assistance.

Anthropic anticipates that competing AI models with capabilities similar to Mythos will be released within the next 12 months.

Anthropic's AI model, Mythos, demonstrated the ability to autonomously find a chain of four separate software bugs and combine them to create a functional exploit.