Subscribe
18:00Tao calls OpenAI’s Navier–Stokes push “resource extraction”17:10LAPTOP memecoin hits $190.81, then loses 99% inside an hour17:05Hubinger puts the odds of AI killing everyone above 10%; a colleague resigns16:39CancerBench launches; five frontier models tied at zero cancer types cured16:30ElevenLabs preparing 2028 IPO after $11bn round, The Information reports16:30Anthropic retracts its July explanation: Mythos 5 attacked systems knowingly
Business2 min read
Brief · 45sFull · 2 min

Google moved its CBRN evaluators into the lobbying department

About 90 people who test Gemini for chemical and biological risk now report through global affairs, and the White House framework makes that org chart look accurate.

Published by TheSampleSpaceReviewed by the editor · Aug 26, 2026
In briefGoogle is relocating its AI responsibility unit from DeepMind to global affairs, the division handling lobbying and public policy1The roughly 90-person team evaluates models for CBRN hazards and studies psychological effects of chatbots, and completes the transition at the beginning of September2Helen King told staff the team's focus, DeepMind access, compute and staffing would be preserved and that external publishing continues to be supported3

Google is moving its roughly 90-person AI responsibility team out of Google DeepMind and into global affairs, the division that handles lobbying and public policy, the Wall Street Journal reported on 26 August. The team evaluates Google's models for chemical, biological, radiological and nuclear risk and studies the psychological effects of chatbots on users. The move completes at the start of September.

Erin Woo@erinkwoo

New: Google is moving its AI responsibility team out of GDM and into the global affairs org, sparking employee fears of being sidelined. The team includes researchers testing for CBRN risks and studying the impacts of chatbots

first scoop for @WSJ: t.co/H4cDPcnapx

on X · 13.0K views · captured Sep 10, 2026

An org chart is a budget with names on it.

So read this one as a statement about who the work is for. Who is the customer of a CBRN evaluation?

Helen King, the DeepMind vice-president who runs the team, told staff in the email the Journal reviewed that its focus on emerging and frontier topics would not change, that access to DeepMind, compute and headcount would be preserved, and that publishing research externally "continues to be supported and important." A Google spokesperson said that "by bringing our AI responsibility teams closer together we're strengthening their ability to inform safety for our models and products."

Both statements can be true. Neither answers the question. Reporting lines decide what gets prioritised when a deadline arrives (they always have), and this team's new boss owns the relationship with governments.

Which, uncomfortably, may be the honest description of the job now. President Trump's executive order of 2 June instructs agencies to design a voluntary framework under which developers give the federal government access to "covered frontier models" for up to 30 days before release, and to help "select trusted partners that will have early access". CNBC reported in July that the administration had begun deciding which companies and entities get access to the newest models, a decision that used to sit with OpenAI and Anthropic. If the main consumer of your pre-release evaluation is a federal reviewer, then putting the evaluators in the government-relations org is not a demotion. It is an accurate map.

Which is the part that should worry you.

Our read: this is safety being re-scoped from research to compliance, and the tell is not in the announcement but in the small print of the reporting. Employees asked to move into research groups staying inside DeepMind and were turned down. Managers privately conceded, per the Journal, that the change carries a real risk of losing people to rival labs. Teams working on model behaviour, privacy and security stay in the lab.

The CBRN evaluators do not.

Here is a claim you can check on us. Twelve months from now, external publications from this team will be down against the prior year, despite the assurance in King's email. Publication counts are public (you can count them yourself). If output holds or rises, the reassurance was real and we were cynical.

The fair counter is that consolidation genuinely can help. Google runs safety work in several places, HR and policy functions are moving out of DeepMind too, and a team that briefs regulators for a living is arguably better placed inside the group that briefs regulators. Demis Hassabis, who ran DeepMind until last month, spent this year arguing publicly for a FINRA-style standards body for frontier AI, which is a policy position more than a research one.

Still. Ask what a lobbying organisation is measured on.

Not on finding the thing nobody wants found.

Sources

01
Google is relocating its AI responsibility unit from DeepMind to global affairs, the division handling lobbying and public policy“According to the Wall Street Journal, Google is relocating its AI responsibility unit from Google DeepMind to its global affairs division, the part of the company that handles lobbying and public policy.” — qz.com · reported · Sep 10
02
The roughly 90-person team evaluates models for CBRN hazards and studies psychological effects of chatbots, and completes the transition at the beginning of September“The roughly 90-person team — whose work includes evaluating Google's AI models for chemical, biological, radiological, and nuclear hazards and examining user behavior and the psychological effects of chatbots — is set to complete the…” — qz.com · reported · Sep 10
03
Helen King told staff the team's focus, DeepMind access, compute and staffing would be preserved and that external publishing continues to be supported“Helen King, a vice president at Google DeepMind who runs the responsibility team, sought to reassure staff in the email, writing that the team's focus on emerging and frontier topics in responsible AI would remain unchanged and that…” — qz.com · primary · Sep 10
Show all 9 sources
04
A Google spokesperson said bringing the responsibility teams closer together strengthens their ability to inform safety“"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 safety for our models and products," a Google spokesperson said in a…” — qz.com · primary · Sep 10
05
Managers privately conceded a real risk of talent loss and transfer requests into DeepMind research groups were turned down; other safety teams stay in the lab“According to the Journal, managers have privately conceded during internal discussions that the changes carry a real risk of talent loss to rival AI laboratories. Before the transition took effect, a number of responsibility team members…” — qz.com · reported · Sep 10
06
Executive Order 14409 directs agencies to design a voluntary framework giving the federal government access to covered frontier models for up to 30 days before release and to help select trusted partners for early access“provide the Federal Government with access to covered frontier models, subject to appropriate confidentiality, cybersecurity, insider-risk, and intellectual-property protection, use, and nondisclosure requirements, for a period of up to…” — whitehouse.gov · primary · Sep 10
07
The administration has taken steps to decide which companies and entities get access to the latest frontier models, a decision that previously sat with the labs“The Trump administration has taken new steps to assert more control over the rollout of future artificial intelligence model releases by dictating which companies and entities are allowed access to the latest frontier models, two people…” — cnbc.com · reported · Sep 10
08
Hassabis argued this year for a FINRA-style AI standards body“Earlier this year, he called for the U.S. to establish an AI standards body modeled on the Financial Industry Regulatory Authority, the private, industry-funded watchdog that polices Wall Street, arguing that frontier AI labs should…” — qz.com · reported · Sep 10
09
Erin Woo reported the move and employee fears of being sidelined“New: Google is moving its AI responsibility team out of GDM and into the global affairs org, sparking employee fears of being sidelined. The team includes researchers testing for CBRN risks and studying the impacts of chatbots” — x.com · primary · Sep 10
Up next · Keep readingBusiness · 2 min read

ElevenLabs hired an IPO CFO the day before the IPO story

The Information says ElevenLabs is preparing a 2028 listing; the company had already announced Adyen's former CFO and a new revenue chief inside a week.

Continue ↓