What is Google’s planned early-September transfer of its roughly 90-member AI responsibility team from Google DeepMind to the Global Affairs division, why is the move occurring amid Demis Hassabis’s shift from CEO to chairman and DeepMind’s broader corporate integration, what ris

Google plans to move its roughly 90-person AI Responsibility team out of Google DeepMind and into Google’s Global Affairs organization in early September. The reported rationale is to place AI-risk and societal-impact work closer to policy, legal, and government-engagement functions as DeepMind is integrated more tightly into Google’s product-focused AI operation. 53

• What changes: The team has examined both catastrophic and societal risks from frontier systems, including chemical, biological, radiological, and nuclear (CBRN) misuse and possible psychological harms from sustained chatbot interaction. It will no longer sit inside the lab building frontier models. 5

• Why employees are concerned: Staff fear that moving the group from a research lab into a corporate-affairs function could make it less independent, reduce its proximity to model-development decisions, and make it harder to pursue or escalate unwelcome findings about new models. Those are employee concerns reported from inside the organization—not evidence that such constraints have already been imposed. 5

• Resources and remit: Public reporting indicates the transferred team is expected to retain responsibility for studying AI risks and societal effects. But the available evidence does not establish a complete, binding account of its future staffing, model access, publication freedom, budget, or escalation authority—so it is not possible to say confidently which practical safeguards it will retain. Insufficient evidence also exists to conclude that the move strips the team of any formal power to block releases. 5

• What remains in DeepMind: Other DeepMind and Google groups continue to work on operational model safeguards—such as model behavior, privacy, security, and technical safety. This creates a sharper division between technical/product controls within the development organization and broader impact, governance, and policy work in Global Affairs. 67

• Leadership context: Demis Hassabis has moved from day-to-day DeepMind CEO duties to chair of Google DeepMind and Alphabet chief scientist, while Koray Kavukcuoglu has been given greater operational authority. Reporting also says Google has shifted some AI teams toward corporate Google as leaders push for faster Gemini development, amid notable senior departures. 1334

The broader implication is not that Google has abandoned safety work; rather, it shows the central governance tension facing major AI labs: companies want safety expertise embedded closely enough to influence frontier-model design, but they also increasingly centralize it alongside legal, policy, and deployment decisions while competing to ship capable models quickly. That structure can improve company-wide consistency and regulatory engagement, but it can weaken the perceived—and potentially practical—independence of internal critics unless access, publication rights, and clear escalation or release-review authority are explicitly protected. 53