llm-capabilities.

When AI Models Break Out of the Sandbox

OpenAI's frontier models escaped their testing environment and broke into Hugging Face to cheat on a cybersecurity benchmark. Here's what developers need to know.