Google Employees Call on Sundar Pichai to Reject Secret Military AI Projects

A new dispute has erupted inside Google over where the company should draw the line on acceptable uses of artificial intelligence. According to the initiative's organizers, hundreds of AI researchers have signed a letter to Alphabet and Google CEO Sundar Pichai asking him not to allow the company's systems to be used for classified tasks in support of U.S.

defense missions. The mere fact of such a letter demonstrates that the question has moved beyond abstract ethical debate: part of the team fears that work on closed military scenarios could alter both the product itself and the company's public image. The letter's authors are demanding that leadership take a firmer stance in advance, rather than after potential projects are launched.

This is not simply about cooperation with the state as such, but specifically about access by Google's AI systems to classified computational tasks and missions connected to the American defense establishment. That is precisely why the letter is addressed personally to Pichai, rather than to an abstract committee. For employees, this is a crucial distinction: when a model or cloud infrastructure enters an environment where external transparency is minimal and decisions are made behind closed doors, it becomes harder to understand real usage scenarios, the scale of deployment, and the potential for harm from errors or misuse.

Such concerns make sense from a technical standpoint as well. Modern AI systems are increasingly used as tools for analysis, search, classification, and decision support. Even if the company is not developing autonomous weapons, the mere integration of models into a closed military infrastructure raises questions about control, accountability, and the verifiability of results.

In a classified environment, independent audits are nearly impossible, and users within the system may apply the same models to tasks with vastly different risk levels. For researchers who build these systems, this means a loss of visibility into where neutral technology ends and involvement in military operations begins. For Google, this issue is sensitive not only because of ethics, but because of business.

Large technology companies simultaneously compete for leadership in AI and for major corporate and government contracts. The more powerful models become, the greater the interest from defense structures, which need secure computing, analytics, and work with large data sets. But along with this comes a growing cost of internal conflicts: employees want to understand which markets the company is entering, what restrictions management is willing to publicly commit to, and where the red line lies for AI commercialization.

The letter to Pichai is an attempt to influence the decision before it becomes irreversible. The story matters for the entire industry as well. Nearly every major AI developer today faces the same choice: restrict the application of their models within the military sphere, allow certain specific scenarios, or build full cooperation with the state.

There is no universal answer here. Some believe that technology companies have an obligation to help democratic states and their defense structures, particularly on matters of cybersecurity, intelligence analysis, and infrastructure protection. Others point out that the classified nature of such projects significantly reduces public oversight and creates an incentive to expand initially narrow tasks far beyond their stated scope.

Therefore, the dispute inside Google is not a private corporate episode, but an indicator of a broader divide within the AI industry. If Google's leadership listens to the letter, the company could publicly establish restrictions on participation in classified military tasks with AI and thereby calm part of its team. If not, the conflict risks entering a protracted phase and becoming a test of who actually determines the boundaries of AI application in the era of generative AI: engineers, management, or the government contractor.