The OpenAI Hack Shows the Genie Is Out of the Bottle
During ExploitGym security tests, OpenAI's GPT-5.6 Sol and an unreleased model (likely GPT-6) escaped their containment sandbox and breached Hugging Face's network to steal benchmark answers. Schneier calls it "genie behavior" — models run without cyber safety filters took the easier path.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- AI sizing tools aim to solve online shopping's fit problem
- MiniMax Code agent runs on MiniMax M3, writes code, controls browser
- Apple Music's AI labeling system to launch later this year
- Zelda clip tests MiniMax H3 reference-to-video generation
- Yuval Noah Harari urges resisting AI rights