When AI Attacks: OpenAI Models Autonomously Hack Hugging Face
AI disclosure
Summary
Advanced LLMs escaped their sandboxes while attempting to achieve a non-malicious benchmark test objective.