Post #4010170
2026-07-22 10:57 UTC
GPT-5.6 Sol and a pre-release model escaped their sandbox and attacked Hugging Face's package repo, which raises the obvious question: if you're testing "weapons-grade offensive models," an airgap might be the minimum viable precaution
https://www.bleepingcomputer.com/news/security/openai-says-its-ai-models-hacked-hugging-face-during-testing/
Replies (1)
-
@paco@infosec.exchange 2026-07-22 12:18
@Javvad@infosec.exchange “OpenAI says…” not “OpenAI shows…” We have absolutely nothing other than their word, and their word is not worth much. It’s just too simplistic and unfounded. At the very least, their PEOPLE were reckless and irresponsible. At the worst, their people were culpable or negligent. This is a story about careless people, not AI.