Google just shifted the roughly 90 people who screen Gemini for bioweapon and psychological risks out of DeepMind and into the arm of the company that lobbies governments.
The Wall Street Journal reported on August 26 that Google was moving its "AI responsibility" team out of Google DeepMind and into Google's global affairs division, the unit that handles policy and lobbying rather than product safety. The transition finished on September 1. This is the team that evaluates Gemini and Google's other models for chemical, biological, radiological and nuclear risk before they ship, and that studies how chatbots affect users psychologically. Now it sits inside the same organization that argues Google's case to regulators in Washington and Brussels.
That's a real change in reporting lines, not a rebrand. Helen King, the DeepMind vice president who runs the team, told staff in an internal email that their focus on "emerging and frontier topics in responsible AI" wouldn't change, and that the group would keep publishing research externally and keep its access to DeepMind's compute and researchers. She framed it as part of a broader shift: DeepMind moving toward a more conventional corporate structure, with HR and policy functions also getting pulled into Google's central organization.
Staff aren't fully buying it. Several members of the responsibility team applied to transfer into research groups staying inside DeepMind before the September 1 deadline. Google turned down every one of those requests.
Why the timing looks even worse #
Google's official line is organizational tidiness. Policy, HR and other support functions are being centralized as DeepMind sheds the structure of a semi-independent lab and becomes just another part of Google. But the practical effect is that the people with authority to hold back a Gemini release over a CBRN finding no longer sit inside the building where the model gets built. They sit inside the division that manages Google's relationship with governments and the public. That's the whole shift, in one move. The Journal's reporting found that Google's own managers have privately conceded the move risks pushing safety researchers toward rival labs. That's an unusually candid admission for a company to make about its own reorganization.
Google DeepMind's Safety Team Warns Job Seekers Its Own AI Hiring Filters Misfire Bloomberg reports that Google DeepMind's AGI Safety and Alignment Team tells job applicants to submit a special workaround form because it doesn't trust Google's own AI hiring filters. The twist: Google markets that same AI screening technology to corporate clients as a selling point. - AI hiring bias in recruiting tools - why AI filters reject qualified candidates
The timing doesn't help the optics. The shift lands the same month Demis Hassabis stepped down as DeepMind's CEO, a move announced on August 5. He's chairman of DeepMind now, and chief scientist of Alphabet, focused on AGI strategy rather than day to day operations. Koray Kavukcuoglu, DeepMind's former chief technology officer, took over as senior vice president running the lab. He reports directly to Sundar Pichai. Jeff Dean, Google's longtime chief scientist, left around the same time to start a new company that Google is backing financially.
Three departures in one summer. All pointing the same direction: less independence for DeepMind, more direct control from Google's core leadership.
The race Google can't slow down for #
None of this happens in a vacuum. OpenAI and Anthropic are both racing to ship frontier models faster than Google can evaluate them, and every extra week spent on a CBRN review is a week a rival might use to grab the headline instead. Frankly, folding the safety check into the same division that lobbies against strict AI rules sends a specific signal about where Google thinks the power sits right now, not with the researchers who might slow a launch, but with the people paid to manage how a launch gets received.
Google didn't say how the model evaluations themselves will change day to day. King's email promised continuity, not independence, and those aren't the same thing. A team that used to sit next to the people building Gemini now has to route its concerns through a division that answers to Google's policy priorities first.
The 90 people affected already know how this usually goes. So did the researchers who asked to stay and got told no.
Also read: China's AI Token Consumption Set to Hit 100 Quadrillion in 2026 • Jacob Coxon Says His Former Anthropic Colleagues Are Genuinely Frightened of AI • Obama Tells Democrats AI Is Moving Fast and Turning Dangerous
Demis Hassabis wants a Wall Street style referee for frontier AI models Google DeepMind CEO Demis Hassabis is calling for a FINRA-style independent body to test frontier AI models for cyber, bio, and deception risks before release. He wants it running by year-end, even as the Trump administration has ruled out formal AI regulation. - frontier AI regulation framework proposal - AI industry regulator like FINRA
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.