Google moves 90-person AI responsibility team out of DeepMind amid reshuffle

VP Helen King informed employees by email that the AI responsibility team — roughly 90 people — will move from DeepMind into Google's wider global affairs organization as part of bringing DeepMind 'more closely into the company structure.' Staff raised concerns that relocating the responsibility function away from the lab could reduce independent research and cut off visibility into Gemini model development.

The reorganization coincides with a leadership shakeup: Demis Hassabis became chairman and chief scientist, with Koray Kavukcuoglu taking over as lab lead. The timing is conspicuous — the change centralizes control just as competitors are publicly committing to MORE independent oversight via Amodei's embedded-evaluator proposal, making Google look like it's moving in the opposite direction on safety governance.

Adding to the intrigue, Gemini 3.5 Pro remains conspicuously unreleased, fueling speculation (Mashable's 'the case of the missing AI model') about whether Google is behind or deliberately holding back. Ethan Mollick noted that if Google returns to the frontier, it would reshape the dynamic as 'a large, public, regulated entity with different goals than Anthropic or OAI.' For now, the org change reads as Google prioritizing integration and speed over the arms-length safety structure it had built.