Did you read any of the METR report about the hugging face incident? This exact type of behavior was predicted many years ago by researchers. It isn't hard to

Did you read any of the METR report about the hugging face incident? This exact type of behavior was predicted many years ago by researchers. It isn't hard to see the trendline of reward hacking and other misaligned behavior over the past couple of years. Especially the past 6 months. The current safety posture is quite poor, to say the least. To be honest, if you don't much experience with or haven't read extensively about ML training and reinforcement learning, then you'll have a hard reasoning accurately about these scenarios. The arguments are not that complicated, but they take us to places that are fairly novel. One may tempted to naively dismiss them out of hand, which is a mistake. These systems are new, their behavior is extremely complex, and they perform actions increasingly far beyond those of any computer program in the pre-LLM era. We are in a new world that requires careful evaluation. Link: https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/ https://metr.org/blog/2026-08-26-openai-hugging-face-inciden...