THE SHORT VERSION
● Sofia Liguori — research engineer, resigned this week
● Andreas Kirsch — researcher, left earlier in October
● The objection — Google's visibility into and control over Pentagon use of its AI
● Google's answer — public-sector contracts go through compliance reviews
Two researchers have left Google DeepMind this month over the same issue. Research engineer Sofia Liguori resigned this week, and researcher Andreas Kirsch left earlier in October. Both raised concerns about Google's visibility into, and control over, Pentagon use of its AI. Google's response is that its public-sector contracts go through compliance reviews.
The objection is narrower than the headline suggests, and that is what makes it interesting. This is not the 2018 Project Maven argument about whether Google should take military work at all. It is about what happens after the model ships: once a system is delivered to a defence customer, how much does the vendor actually know about how it gets used, and what can it do if the answer is something it objects to? A compliance review happens before deployment; the concern is about the period afterwards, which has no clean technical fix.
Every frontier lab now has defence revenue or is pursuing it, and the standard public answer is a usage policy plus a review process — the answer Google has given here. Two researchers deciding that assurance was insufficient is a signal about how it reads from inside the building. Worth watching whether it stays at two: individual resignations are individual decisions, and what changed outcomes at Google in 2018 was a letter with thousands of signatures, which has not appeared. Neither researcher has alleged a specific misuse and no contract has been named, so anything stronger than "a disagreement about oversight sufficiency" is inference rather than reporting.
Sources
The Neuron daily digest