[dupe] https://news.ycombinator.com/item?id=48997548

“OpenAI confirms their agent harness and security failures allowed a prompted AI to exceed permissions.” Would be a more descriptive title without OpenAI FUD.

It seems that it all is going in the way everybody expected.