I would love to see media outlets be more careful in how they describe AI tools attacking other systems. They’re not “rogue”. They are malicious. And their authors are still responsible for them. If they’ve acted irresponsibly with an execution environment with too few controls, then it’s the humans in charge of creating that environment who are ultimately to blame. https://www.cbc.ca/news/canada/openai-university-toronto-rogue-agents-link-shortener-ai-9.7349607 Code doesn’t “escape”. It only has the access we give it.