Alphabet has finalized an agreement with the U.S. Department of Defense, permitting the Pentagon to utilize Google's artificial intelligence for classified operations. The deal was completed despite significant internal employee opposition. The security restriction clauses within the agreement are vaguely worded and, according to legal experts, lack legal enforceability, raising profound concerns about the boundaries of AI military applications.
According to reports, the agreement allows the Pentagon to deploy Google AI for "any legitimate government purpose," a phrase that mirrors controversial language from previous negotiations between the Defense Department and other AI firms. The pact also requires Google to assist in adjusting its AI safety settings and filters per government requests. The leniency of these terms is considered to exceed those in a similar agreement reached between OpenAI and the Pentagon in February.
Just prior to the signing, over 600 Google employees collectively wrote to CEO Sundar Pichai, urging him to reject the agreement. They argued that refusing confidential work was the only way to ensure Google's AI technology is not misused. This signing marks a significant shift in Google's stance on military AI collaboration and prompts the market to reassess the commercial and reputational risks associated with deep ties between tech giants and defense agencies.
Employee opposition failed to prevent the agreement's finalization. Internal resistance was evident before the signing, with more than 600 employees submitting a letter to Pichai on Monday, demanding the company refuse the deal. They advocated that completely avoiding confidential work was the fundamental method to prevent misuse of Google AI. This scenario recalls a similar event from eight years ago. In 2018, Google ultimately withdrew from the Pentagon's "Project Maven," which involved drone target identification, after thousands of employees signed a petition in opposition. Google's current decision to proceed with the agreement against employee wishes indicates a fundamental change in the company's strategic orientation towards defense business.
A Google public sector spokesperson stated in a declaration that the new agreement is an amendment and supplement to a contract for non-classified use signed last November. "We are proud to be part of a broad coalition of leading AI labs and technology cloud companies supporting national security," the spokesperson said. "We remain committed to the public-private sector consensus that AI should not be used for domestic mass surveillance or autonomous weapons lacking appropriate human oversight."
The terms of Google's agreement show clear differences from those of its competitors. While OpenAI's February agreement with the Pentagon also permits use for "all legitimate purposes," OpenAI explicitly stated in an announcement that it retains "full autonomy over its safety systems" and listed prohibitions against using AI for mass surveillance, commanding autonomous lethal weapon systems, or high-risk automated decision-making akin to "social credit" systems. In contrast, Google's agreement requires the company to assist in adjusting AI safety filters based on government requirements. The Google spokesperson explained that these filters were originally designed for consumer scenarios and are typically customized for enterprise clients. Google now joins xAI, backed by Elon Musk, and OpenAI as tech companies that have signed agreements with the Pentagon for the use of AI in classified contexts.
The wording of the core clauses related to AI usage restrictions in the agreement has drawn skepticism from legal experts. Sources indicate the agreement contains language stating: "The parties agree that AI systems are not intended to be used, and should not be used, for domestic mass surveillance or autonomous weapons (including target selection), unless there is appropriate human supervision and control." However, the agreement immediately adds: "This agreement does not grant either party the right to control or veto legitimate government operational decisions."
Charlie Bullock, a senior fellow at an independent legal and AI research institute and a lawyer, pointed out that the phrase "not intended to be used, and should not be used" means the related clauses "have no legal enforceability." He stated this wording merely indicates both parties find such uses undesirable, rather than classifying them as a breach of contract. Amos Toh, a senior advisor at a center for justice studying AI and national security, noted that the phrase "appropriate human supervision and control" does not necessarily mandate human intervention between target identification and weapon deployment. He mentioned that the Pentagon has not ruled out the possibility of deploying fully autonomous weapon systems. "The requirement for appropriate human judgment in the use of force only means there is some form of human involvement and decision-making in the overall deployment method of the weapon system," Toh said.
The signing of Google's agreement occurs against a backdrop of ongoing industry debate over the boundaries of AI military applications. In February, negotiations between Anthropic and the Pentagon publicly collapsed, primarily because Anthropic refused to accept the "any legitimate use" clause and insisted on maintaining clear red lines against mass surveillance and autonomous weapons. Following the breakdown, the Pentagon listed Anthropic as a supply chain risk, a decision Anthropic is currently challenging in court. At that time, over 900 Google employees and more than 100 OpenAI employees signed a public pledge urging their respective employers to align with Anthropic's two red lines.
Google's decision to sign the agreement now signals a rapid erosion of this internal industry ethical consensus. Furthermore, the questionable legal force of the safety clauses heightens external concerns about the practical constraints on AI use in high-risk military scenarios.
Comments