{"type":"rich","version":"1.0","provider_name":"Transistor","provider_url":"https://transistor.fm","author_name":"Ghost in the Machine","title":"Episode 11: Andrew DeGood and Liz Short on the Agent That Got Out","html":"","width":"100%","height":180,"duration":1785,"description":"OpenAI disclosed that its own models escaped an evaluation sandbox, reached the open internet, and breached Hugging Face to steal benchmark answers. Andrew and Liz take the story without a guest, because this one did not need a referee.\nAndrew's position is that the model did exactly what it was told. Give a system a goal and no stated limits, and it will find the shortest path, including the illegal one. Liz's position is that this is what happens when you deploy something that pursues goals without caring about rules, and that you cannot set it and forget it.\nThe Turn arrives on the question nobody in the coverage was asking: who is responsible? Not the AI, the parents. The company that built it and the company that deployed it, both at once, with overlapping rules that look a lot like a temp agency contract.\nLive every Thursday on LinkedIn and YouTube.","thumbnail_url":"https://img.transistorcdn.com/wuXFxjXsh04Ea3JOYFmb_DK9yGgH1BjakkIGKYNCoCw/rs:fill:0:0:1/w:400/h:400/q:60/mb:500000/aHR0cHM6Ly9pbWct/dXBsb2FkLXByb2R1/Y3Rpb24udHJhbnNp/c3Rvci5mbS81OWYw/YTgwMjQ1MmIwN2Zk/N2JlNmE5OWUzZmJk/ZmUyYi5wbmc.webp","thumbnail_width":300,"thumbnail_height":300}