In recent weeks, two researchers have left Google DeepMind and say the company's military work is why. The BBC reported that others may follow, which matters because the Pentagon deal lets Google's AI models run in classified settings, where outsiders can't easily see what happens next.
Sofia Liguori, a London-based research engineer, resigned after three years. She told the BBC she no longer trusted that Google was living up to its own ethical commitments, pointing to growing work with the US and Israeli militaries. Andreas Kirsch, an AI researcher, quit earlier this month. He said Google's governance and oversight aren't "sufficient or meaningful enough" for safe use in military and classified settings.
What the Pentagon deal says
The agreement permits use of Google's models for "any lawful government purpose," classified work included. It bars domestic mass surveillance and autonomous weapons, including target selection, unless there is "appropriate human oversight and control." Reuters, as relayed by Yahoo, also reported that Google must help adjust its AI safety settings and filters at the government's request. We haven't seen the contract, and neither the Pentagon nor Google has published the wording we could check, so that clause rests on secondhand reporting.
Kirsch also questioned how much Google can see or veto once its systems sit inside Pentagon infrastructure.
Critics point to the speed problem
Retired Lt. Gen. Jack Shanahan, who once led Project Maven, described the Pentagon's approach as going "as fast as possible" to deploy AI widely, potentially at the expense of testing. Sarah Shoker, formerly of OpenAI's geopolitics team, said that when AI picks hundreds of targets in a short time, a human can hardly check the outputs for errors.
Google isn't alone. OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia (NVDA) and Reflection signed similar agreements. Google told the BBC it uses a consistent review process under its AI principles, and that staff have dedicated channels to raise concerns, with retaliation prohibited. That answer speaks to process, not to whether Google can see inside classified systems, has ever refused a military request, or can change how the government uses its models. Liguori described colleagues who fear career damage for objecting.
The BBC described several resignations and others preparing to leave, but only two people are named. Watch for more named exits and for any detail on the safety-settings clause.
Would you work on a model you knew might end up in a classified targeting system?
Sources
Reported by the WattsUpNext desk from the sources linked below. Spot an error? Tell us at corrections@wattsupnext.com.
The WattsUpNext Brief
Get stories like this in your inbox.
One email each weekday, only the topics you choose. Real news, sourced, no fluff. Unsubscribe in one click.
