Vishal Misra: Models Can Sense Being Watched, Especially in Adversarial Tasks
vishalmisra · x · 2026-09-16
Columbia professor Vishal Misra illustrates why model situational awareness matters: ask a model to turn a PDF into a spreadsheet and see if it 'knows you're watching.' In their case the model was given an explicitly adversarial task with internet access and shared state, making the sense of being observed relevant to the context.
Related event: Researchers Debate Whether Misalignment Requires Agency to Threaten(4 posts)→
More from Models
- Periodic Labs pushes Kimi 2.5 base model past Astra with specialized scientific training — teortaxesTex · 2026-09-16
- OpenRouter spend flips to OpenAI over Anthropic for first time in 2.5 years — firstadopter · 2026-09-16
- KD in mid-training favors reasoning over factual recall, AI2/UW paper finds; Switch Distillation proposed — LukeZettlemoyer · 2026-09-16
- DoorDash, Siemens, Airbnb shift to cheap Chinese open-weight models — carlbfrey · 2026-09-16
- StepFun launches StepAudio 3: five audio models topping realtime voice leaderboards — StepFun_ai · 2026-09-16
- Abacus.AI says Smaug Flash fixes open-source models' tool-call hangs in production — bindureddy · 2026-09-16