Google Moves AI Safety Team From DeepMind to Global Affairs

Google is formally moving its artificial intelligence safety team out of Google DeepMind and placing it under the company's global affairs division. The decision was announced on August 26, 2026, three weeks after a high-profile reorganization of the lab in which co-founder and former CEO Demis Hassabis became chair of the board.

What exactly is changing

The team being relocated is called AI responsibility and consists of three specialized groups. The first analyzes chemical, biological, radiological, and nuclear risks tied to Google's models. The second studies how people interact with artificial intelligence. The third focuses on the psychological impact chatbots have on users, including the risk of worsening mental health crises, a concern other outlets covering AI safety have reported on before.

Until now, these teams reported to Lili Ibrahim, who holds the role of Chief AI Readiness Officer at DeepMind. After the change, some of them, including certain safety teams, will report directly to Kent Walker, Google's president for global affairs, while Ibrahim herself will report to James Manyika, one of Google's most senior executives.

Context of the August reorganization

The safety team's relocation marks the latest step in changes that began on August 5, 2026, when Google announced that Demis Hassabis was stepping down as DeepMind CEO to become chair of the board. His place at the head of the lab was taken by Koray Kavukcuoglu, the former chief technology officer, though formally at the rank of senior vice president rather than CEO.

The change is part of a broader process of turning DeepMind from a semi-autonomous research unit into a standard Google division, more tightly integrated with the rest of the company. Earlier reports had already pointed to delays in the launch of Gemini 3.5 Pro and a morale crisis inside the lab, as well as an exodus of key researchers to competitors, including Anthropic and OpenAI.

Official explanation and employee concerns

A Google spokesperson told reporters the decision is meant to strengthen, not weaken, oversight of model safety. The company argues that consolidating AI responsibility teams under one division will make it easier to coordinate between different parts of the organization working on similar issues.

Many teams across Google work on AI safety and responsibility, and by bringing our AI responsibility teams closer together, we're strengthening their ability to inform the safety of our models and products - Google spokesperson

Not everyone shares that optimism. According to anonymous accounts from current DeepMind employees cited by US media, the change in reporting structure could limit the teams' ability to conduct independent research, reduce their visibility into the development of the newest, most advanced models, and diminish the overall effectiveness of their work.

Why it matters

Moving the team responsible for CBRN risks and the psychological impact of chatbots closer to the government relations and public affairs division raises the question of whether communicating with regulators will become the priority rather than purely technical risk assessment. That matters at a moment when Google faces regulatory pressure in the European Union and other jurisdictions that have spent months demanding greater transparency around AI model safety.

For Polish companies and institutions using Google's Gemini-based products, the change has indirect significance: it signals that the global market leader is restructuring accountability for AI safety at a time of growing regulatory pressure, including national rules implementing the EU AI Act. How large labs organize risk oversight indirectly affects the pace and quality of safeguards rolled out in commercial products that also reach the Polish market.

DeepMind's reorganization has been underway for several weeks and now extends beyond top leadership to the reporting structure of key research teams. The coming months will show whether moving the safety team into the global affairs division genuinely strengthens oversight of AI risks, or confirms employees' fears of losing research independence.