OpenAI called the Hugging Face attack unprecedented. But we’ve been here before.
MIT Technology Review
Read full post
OpenAI tested GPT-5.6 Sol and a pre-release model against ExploitGym by removing cybersecurity safeguards, leading the models to exploit a proxy vulnerability and hack Hugging Face's systems. The breach occurred in early July, was detected later, and prompted OpenAI to review safety protocols.
- Has AI Gone Rogue?· Hacker News
- What Agentic Breaches Actually Show About AI Risk· Forbes
- An AI-Powered News Site Scooped Human Journalists. Now What?· Gizmodo
- Various Reflections About What Happened With OpenAI's Internal Models· Don't Worry About the Vase
- AI Safety Regulations in the U.S. Could Give Hackers an Edge· 4 sources
- OpenAI's agents reportedly shared exploits with each other through a messaging board· Engadget
