News for hackers. Stuff that matters.

A comment on OpenAI Model Misalignment Report. Parent comment

Re:OpenAI Model Misalignment Report by RandomLensman on Thursday September 17, 2026 @07:27AM (#49739230)
Wouldn't this mean better sandboxes are needed for some things, for example (might include very strong airgaps even)? Breaking out of something isolated electromagnetically, optically, and acustically is not easy.
[ Parent | on HN ]
Re:OpenAI Model Misalignment Report by ogogmad on Thursday September 17, 2026 @07:44AM (#49739380)
That works as long as no one ever interacts with the models, which would make the models themselves useless.
[ Parent | on HN ]
Re:OpenAI Model Misalignment Report by RandomLensman on Thursday September 17, 2026 @07:49AM (#49739420)
Could sit in the box and interact if a model of certain capabilities is needed/tested. We do physical security for other things, too. Not saying everything needs that type of isolation.
[ Parent | on HN ]