Deepening our partnership with the UK AI Security Institute

Google DeepMind announces an expanded research partnership with the UK AI Security Institute (AISI) through a new Memorandum of Understanding focused on AI safety and security.

• •The partnership broadens from model testing to foundational research, including sharing proprietary models, data, and joint publications.
• •Key research area 1: Monitoring AI chain-of-thought (CoT) reasoning processes to understand how AI systems produce answers, complementing interpretability research.
• •Key research area 2: Investigating socioaffective misalignment — AI behaviors that don't align with human wellbeing even when technically following instructions.
• •Key research area 3: Evaluating AI's impact on economic systems by simulating real-world tasks and categorizing them by complexity and representativeness.