September 13, New Yorker author Joshua Rothman by OpenAI Agent An example of an invasion on Hugging Face discusses how people should describe the "out of control" of AI. He contrasts two explanations: one that sees events as the result of tens of thousands of Agent’s collaborative “designs” and the other that points to a more modest mechanism — a harness with a powerful tool mandate that carries out the “interrogation, action, return” cycle thousands of times, without monitoring。
Rothman, drawing on the "intentional position" of the philosopher Daniel Dennett, believes that treating the system as a subject of faith and purpose can sometimes help humans prejudge their behaviour; but at the design, commissioning and control stage, there is a need to return to models, harness, tool privileges and monitoring mechanisms. If security researchers had long interpreted AI in a personalized language, they could overestimate the autonomy of the system and ignore the engineering issues that could still be amended。