Internal email: Google is moving DeepMind's ~90-person “AI responsibility” team, focused on the risks and societal impact of AI, to Google's global affairs unit
Researchers have raised concerns about the change's effect on their independence and ability to detect threats from newest models
Context & Ripple Effects
Google had already centralized Responsible AI work inside DeepMind in 2024 as it consolidated model-building groups, following the creation of the Google DeepMind super-unit around a more product-oriented AI lab. Its August 2026 leadership reshuffle put DeepMind under Koray Kavukcuoglu, while other teams moved into corporate Google.
Placing the responsibility group in global affairs separates safety and societal-impact research from the frontier-model organization it evaluates. Researchers' stated concern is that the new reporting line weakens their independence and capacity to identify threats in the newest models.
First-order effects
- Google's roughly 90-person AI-responsibility team shifts from DeepMind to global affairs, changing the internal channel through which its risk work reaches model leadership.
- DeepMind loses a dedicated in-house group focused on AI risks and societal effects; the affected researchers have raised concerns about their independence and threat-detection role.
Second-order effects
- Google's model and product teams will have to incorporate safety findings through a corporate-affairs structure rather than a unit embedded in DeepMind, increasing the importance of escalation and decision rights.
- The move puts Google’s external-policy and reputational considerations closer to frontier-model risk research, making the credibility of internal governance a more visible issue for the company.
Third-order effects
- If frontier labs continue moving safety functions into corporate organizations, AI governance may become less a research-lab function and more an operational and public-affairs discipline.
- The reorganization reinforces a broader tension in AI industrialization: firms seeking faster model diffusion must preserve enough independent challenge to make internal risk review credible.
The trend: Frontier AI labs are institutionalizing governance inside corporate structures as model development, product deployment, and public-policy exposure converge.