A powerful new AI model named Mythos demonstrated alarming capabilities, including finding and exploiting previously unknown vulnerabilities. Due to these risks, its developer, Anthropic, opted not to release the model to the general public.