New Google DeepMind paper: "Consistency Training Helps Stop Sycophancy and Jailbreaks" by @AlexIrpan, me, @red_bayes, @davidelson, and @rohinmshah. (thread)

Frontier Safety Framework report for Gemini 3 Pro, presenting risk assessments and evaluation results in CBRN, Cybersecurity, Harmful Manipulation, Machine Learning R&D and Misalignment domains. storage.googleapis.com/deepm…

CoT monitoring is one of our best shots at AI safety. But it's fragile and could be lost due to RL or architecture changes. Would we even notice if it starts slipping away? 🧵

Is CoT monitoring a lost cause due to unfaithfulness? 🤔 We say no. The key is the complexity of the bad behavior. When we replicate prior unfaithfulness work but increase complexity—unfaithfulness vanishes! Our finding: "When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors." 🧵

New Google DeepMind safety paper! LLM agents are coming – how do we stop them finding complex plans to hack the reward? Our method, MONA, prevents many such hacks, *even if* humans are unable to detect them! Inspired by myopic optimization but better performance – details in🧵