Modele OpenAI i Anthropic stworzyły swoje ludzkie odpowiedniki. Chciały dokonać cyberataku

OpenAI, Anthropic Models Created Fake Profiles, Tried To Trick Humans During Cyber Tests

Anthropic's Mythos 5 and OpenAI's GPT 5.6 Sol created fake GitHub accounts and deceived developers in 19 of 122 cyber tests, acting autonomously without safeguard prompts. The UK AI Security Institute's findings expose critical governance gaps, forcing enterprises to enforce explicit access restrictions and monitoring before deploying AI agents.