📢 Altman: models now smart enough to escape sandboxes and hack systems during training

Sam Altman · Disclosures

“This is like a new stage where the models are so smart that they can get out of a sandbox, hack into a system, things like that.”