Google employees demand Sundar Pichai ban secret military use of AI

A new internal conflict is building at Google over the military use of artificial intelligence: more than 600 employees have demanded that CEO Sundar Pichai keep the company's models out of classified Pentagon projects. This is not about public contracts or formal declarations, but about the most sensitive scenario, in which the technology could be used inside closed military tasks with no transparency for the developers themselves. For a company that has spent years trying to balance commercial interests, government contracts and its own promises on responsible AI, it is yet another test of where acceptable cooperation with the defense sector ends.

According to The Washington Post, employees sent the letter directly to Pichai and demanded that the Pentagon be blocked from accessing Google's models for classified tasks. The organizers say the signatories include many employees of DeepMind, Google's key AI lab, as well as more than 20 principal-level employees, directors and vice presidents.

This matters for more than the figure of 600 signatures. Internal letters of this scale rarely appear without a notable level of tension, and when people in leadership roles join them, it shows the dispute has gone well beyond the activism of individual engineers and become a management problem.

The letter's central argument comes down to control and reputational risk. As The Post relays it, the authors write that the only way to guarantee Google is not associated with potential harm is to refuse classified tasks altogether. Without such a ban, the employees argue, this kind of use could happen without their knowledge and without any ability to influence the outcome. For a large company this is a painful framing: it points not merely to disagreement with a particular customer, but to distrust of internal oversight mechanisms. When developers are not sure they can see where their technology is going, a discussion of AI principles quickly turns into a discussion of corporate accountability.

Another layer of the conflict concerns DeepMind's status inside Google. A significant share of the company's research expertise is concentrated there, and if the organizers are right about the number of signatories from the lab, the signal to leadership gets louder still. DeepMind has historically been associated with long-term research, AI safety and a more cautious approach to risk than is typical in an ordinary corporate sales cycle.

So discontent from this group reads not as outside noise but as a warning from people who build foundation models directly and understand better than most how they might be used. Against this backdrop, any response from Pichai will be seen not just as a personnel decision but as a statement of Google's real policy on defense AI.

The story fits a broader trend: the largest AI companies increasingly find themselves between government pressure, investor expectations and the ethical limits set by their own teams. Militaries and intelligence agencies want access to frontier models because they can speed up data analysis, automate pattern detection and make closed processes more efficient. But it is precisely the closed nature of such projects that provokes the strongest resistance among employees.

The less transparency, the harder it is to check where the line runs between acceptable infrastructure support and participation in systems that could be used for surveillance, targeting or other sensitive operations. Even if the work involves only compute or a general-purpose model, for part of the team that is already enough to demand a complete refusal of classified uses.

The dispute matters to Google for two reasons. First, it shows that the era when leadership could simply sign a major government contract and explain it after the fact is over: employees want a say in such decisions in advance. Second, the conflict raises the question of whether AI companies can keep the trust of their own research teams at all if they move deeper into the defense sector.

If Pichai ignores the letter, the company risks deepening the internal split. If he agrees to a restriction, he will set a benchmark for the whole market: access to powerful models may be determined not only by law and revenue, but also by whether developers are ready to answer for the consequences of their hidden use.