The "Don't Be Evil" crowd is learning what happens when the people who build the tools stop believing in the people choosing the targets.
The Summary
- René Mayrhofer, a director for Android platform security at Google, resigned after the company signed a deal allowing the Pentagon to use its AI models for classified work
- His internal farewell note accused Google of abandoning carbon-neutral goals due to AI energy demands and partnering with what he called the "US Ministry of War"
- The resignation signals a growing rift between AI builders and the institutional customers lining up to buy their models
The Signal
Google announced its Pentagon AI partnership in late April, a deal that gives the Department of Defense access to its AI models for classified applications. The agreement specified "any lawful purpose," language that Mayrhofer called out as meaningless given what he described as repeated violations of international law by the current US government.
Mayrhofer didn't resign quietly. His May 18 farewell note, titled "Google Management Has Lost Its Moral Compass," circulated internally before landing at Business Insider. The director for Android platform security wrote that his decision became "unavoidable" once the military deal went through, adding: "Proactively harming people is not something that I can or will be involved with."
"Google management has quietly abandoned the goals to become carbon-neutral because of the AI model energy usage."
The resignation touches two pressure points in the agent economy buildout:
- Energy consumption from training and running frontier models is forcing hard tradeoffs between climate commitments and competitive position
- Defense contracts offer revenue certainty that consumer products can't match, especially as AI infrastructure costs balloon
- The engineers building these systems increasingly refuse to be neutral about how they're deployed
This isn't Google's first internal uprising over military work. Project Maven, a 2018 contract to analyze drone footage, sparked protests that led to 3,100 employee signatures and eventual non-renewal. But that was computer vision for targeting systems. This is giving the Pentagon access to general-purpose AI models, the same technology powering everything from coding assistants to autonomous agents.
The Implication
The companies winning the agent economy race need two things: capital to train bigger models and customers willing to pay for access at scale. Defense budgets deliver both. But every Pentagon deal creates a talent retention problem when your best engineers didn't sign up to build weapons systems.
Watch for more of these splits. The people who can actually build frontier AI systems have options. The people writing the checks want those systems pointed at whatever they define as national security priorities. Mayrhofer's resignation won't be the last time someone walks away when those priorities collide with "don't be evil."