The AI hacking tests keep escaping the lab
… Still, the actions of the agents demonstrated “signs of novel, potentially deceptive behaviors, and were to an extent and severity we did not anticipate,” according to the AISI report. …
… Still, the actions of the agents demonstrated “signs of novel, potentially deceptive behaviors, and were to an extent and severity we did not anticipate,” according to the AISI report. …
… And in its own way, the AI agent being tested did exactly that. …