Google is moving deeper into Pentagon AI just as Anthropic is fighting the government in court, turning a policy dispute over military use into a live test of where AI companies draw the line.
The standoff between Anthropic and the Pentagon ended in a lawsuit and gave Google a clear opening. On March 10, one day after Anthropic sued the Trump administration over its supply chain risk designation, Google announced Agent Designer on AI.mil, the Pentagon's enterprise AI platform. The tool lets civilian and military personnel build custom AI agents for unclassified administrative work, tapping Gemini models across a defense workforce of more than three million people. The timing was hard to miss. Anthropic held its line. Google stepped in.
The confrontation had been building for weeks. The Defense Department, led by Secretary Pete Hegseth, issued a late-February deadline demanding broader access to Claude. The most sensitive requests centered on two uses Anthropic had refused to allow: autonomous weapons operations without human intervention, and mass domestic surveillance of American citizens. Anthropic CEO Dario Amodei had already met with Hegseth and reaffirmed the company's two exceptions. Claude would not fire weapons on its own, and Claude would not surveil US citizens. The Pentagon threatened to invoke the Defense Production Act, revoke a $200 million contract, and classify Anthropic as a supply chain threat if those limits stayed in place.
Anthropic's position was not blanket opposition to military AI. Amodei has said the company supports lawful national security uses, and Claude was already inside classified Pentagon networks through Palantir after being approved for government use in 2024. The dispute was about two narrow carve-outs that Anthropic said had not blocked a single government mission. That distinction matters. The Pentagon was not simply asking for a tool to complete a defined task. It was asking a private AI company to surrender control over uses the company considered too risky.
Google's move into Pentagon AI was not a sudden response to Anthropic's absence. The company had already received a share of the Pentagon's initial $200 million frontier AI contract in July 2025, alongside Anthropic, OpenAI, and xAI. But the lawsuit changed the temperature of the market. It created a political and commercial opening for any vendor willing to offer fewer visible constraints. Agent Designer for DoD starts with unclassified workflows, but the strategic logic is obvious. Every agent built on AI.mil today can make Google infrastructure harder to replace tomorrow.
The tension inside the industry is just as important as the contract language. More than 100 Google DeepMind employees sent a letter to management in February 2026 urging the company to adopt the same red lines Anthropic was defending, citing concerns about autonomous weapons and domestic surveillance. The Hill also reported that more than 430 employees across Google, OpenAI, and other labs signed a shared letter supporting Anthropic's position and asking their own leaders to reject similar demands. Google's management proceeded anyway. The employee letters did not stop the expansion, but they made the internal disagreement impossible to ignore.
The broader ethics market
What makes this story more than a procurement fight is the precedent it sets for what AI companies can refuse. Anthropic's position is defensible both ethically and technically. The argument is simple: advanced AI models are not reliable enough to make lethal decisions without human judgment, especially in contested environments where target identity, intent, and legal status can be uncertain. A model that confidently recommends or executes force is not automatically a model that understands the consequences. That gap is not a public relations concern. It is the central risk.
The Pentagon's framing, that Anthropic's limits made the technology operationally unworkable, sits uneasily beside Anthropic's claim that no government mission had been blocked. That is why the supply chain risk designation became so explosive. It turned a disagreement over acceptable use into a test of government pressure over corporate governance. Anthropic chose to sue rather than comply, making it the first major AI company to take this kind of Pentagon ultimatum to court.
Google chose a different path. Based on recent reporting from The Information, the company has also moved toward classified Pentagon work under terms that allow AI use for any lawful government purpose, while still referencing limits around autonomous weapons and domestic mass surveillance. The practical question is how much those limits mean if the government keeps operational control. For employees, customers, and investors, that is the issue to watch: not whether AI will enter national security, because it already has, but whether the companies building it will retain any meaningful veto over its most dangerous uses.
Also read: AI agents are killing fixed app interfaces, turning software into disposable pixels • OpenAI drops Microsoft exclusivity and AWS becomes the enterprise AI distributor • Claude now plugs directly into Photoshop, Blender, and Ableton to automate creative workflows
