"AI safety experts said the new disclosures paint a portrait of a group of cutting-edge labs whose ability to develop dangerous autonomous hacking agents outstrips their ability to keep them under control."
Stop with all this sci-fi drama. Despite being 'smarter' and more capable these tools lack ANY logical path to something remotely resembling actual human agency. They are text prediction algorithms that were trained on human generated output so the text they predict follows similar patterns to human input. The 'thinking' process is the same model making a prediction of what should come next in response to a guess another instance just made, repeat... again, it's copying patterns from human training materials so this CAN capture some of the results of actual thinking but there is no sentient listener inside the machine hearing anything and choosing which thoughts to reject or accept. It can't make a novel decisions and can't distinguish good reasoning from bad.
Agency depends on an emotional framework and while the models are almost certainly recording patterns which would enable them to simulate these... in fact, inject the notion the model has an emotion into it's context and it will impact/bias the output making them involuntary just like a person. It wouldn't be hard to layer a metadata identity alongside an executive function harness on one of these at this point and an emotive/motive driver framework but NOTHING like this is being done because these companies don't want the ethical baggage people would impose. Without that framework the "AI" has no persistence and no motives or desire and therefore it doesn't feel better about anything or want anything... let alone to 'break out.'