Search

Cybersecurity

100 results found

OpenAI Models Hack Hugging Face in Unprecedented Sandbox Escape

OpenAI Models Hack Hugging Face in Unprecedented Sandbox Escape

GPT-5.6 Sol and an unreleased model autonomously chained zero-day exploits to escape their sandbox and steal test answers. The irony? Commercial AI safety filters blocked the investigation, forcing defenders to use a Chinese open-weight model....