The Culture of Proximity
Since its acquisition by Google in 2014, DeepMind has operated with a degree of semi-autonomy, fostering a unique, research-driven culture. This environment was built on 'lab proximity'—not just physical closeness, but a shared intellectual space where
researchers collaborated intensely. In the early days, AI safety was sometimes seen as a fringe topic, so the team framed its work in the language of machine learning to gain traction. This close-knit structure allowed for a nuanced, culturally embedded approach to responsibility. Ideas were debated, projects were scrutinized by peers in the same room, and a collective sense of caution was part of the lab's DNA. This informal governance was seen by many as a crucial safeguard, ensuring that the profound risks of advanced AI were intuitively understood and managed by the very people building it.
The Shift to Centralised Governance
Recently, Google has moved to more tightly integrate DeepMind into its corporate structure. This includes a significant reorganisation announced in August 2026, which saw co-founder Demis Hassabis move to a chairman role. A key part of this shift involves moving DeepMind's 90-person 'AI responsibility' team out of the lab and into Google's central global affairs organisation, which handles lobbying and public policy. This unit is responsible for studying some of the most serious AI risks, including potential chemical, biological, and nuclear threats, as well as the psychological impacts of chatbots. Google's official stance is that bringing its various responsible AI teams closer together will strengthen their ability to ensure safety across all models and products. This move represents a shift from a bespoke, lab-specific culture to a standardized, policy-driven corporate framework, in line with a global trend toward more formal AI governance.
Fears of Lost Insight and Independence
The reorganisation has sparked anxiety among some DeepMind employees. The core concern is that moving the safety team away from the developers of models like Gemini will diminish their influence and insight. Employees worry that without the day-to-day proximity to the engineers building the AI, it will be harder to identify new risks, conduct independent research, and understand the nuances of the latest models. This sentiment echoes broader concerns raised in open letters from employees at major AI labs, who warn that companies often possess significant non-public information about the risks of their systems and lack strong obligations to share it. The fear is that a centralized policy team, removed from the frontline of research, may be less effective at catching subtle but critical safety issues before they escalate.
The Case for Formal Policy
However, the move toward formal oversight isn't without merit. As AI technology scales and becomes embedded in global products, relying on the informal culture of a single lab becomes difficult. Formal governance structures, like Google's AI Principles first introduced in 2018, aim to create consistency, accountability, and a clear review process for all projects. This includes internal review boards and accountability mechanisms designed to ensure systems are safe, fair, and socially beneficial. Supporters of this approach argue that centralized teams can apply standards uniformly across the entire company, avoiding silos and ensuring that safety isn't dependent on the culture of one specific group. A Google spokesperson stated that the change strengthens the company's ability to inform safety for its models and products by bringing responsibility teams closer together. This aligns with the broader industry and regulatory push for auditable, transparent, and enforceable AI governance frameworks.
A Debate for the Future of AI
The tension at DeepMind is more than just a standard corporate restructuring. It strikes at the heart of a critical question for the entire technology industry: What is the best way to ensure the safe development of artificial intelligence? Is it through the organic, high-context awareness that comes from a tight-knit research community, or through scalable, enforceable policies managed by a central authority? The employees' concerns suggest that these two models may be in conflict. Losing 'lab proximity' could create a gap between the teams developing the technology and those tasked with overseeing its risks. The outcome of this internal debate at one of the world's leading AI labs could set a precedent for how the industry balances the rapid pace of innovation with the profound responsibility of managing its consequences.
