Google DeepMind deepens partnership with UK AI Security Institute
A new memorandum of understanding extends a relationship dating to November 2023 into joint research on chain-of-thought monitoring and AI's labour-market effects.
- Government & policy
- Safety & alignment
- Colour
Google DeepMind signed a new memorandum of understanding expanding its relationship with the UK’s AI Security Institute (AISI), which had begun with pre-deployment testing access in November 2023. The expanded partnership adds shared access to proprietary models and data, joint published research, and collaboration across three specific areas: monitoring models’ chain-of-thought reasoning for signs of deceptive or unsafe planning, “socioaffective alignment” — whether a model’s behaviour can satisfy an instruction while still harming a user’s wellbeing — and simulating AI’s effects on labour markets.
The announcement gave no new testing timelines or model-specific commitments beyond the existing pre-release access arrangement, and read as a formalisation of an already-established relationship rather than a new undertaking. It is one of several bilateral MOUs AISI has signed with frontier labs as the UK sought to maintain a testing role after the 2023 Bletchley Park summit that created it.