In response to OpenAI’s contentious agreement with the US Department of Defense, Caitlin Kalinowski resigned as head of the company’s hardware division.
“It was not an easy decision. Artificial intelligence plays an important role in ensuring national security. But surveillance of Americans without a court warrant and lethal autonomy without human approval are issues that deserve more scrutiny than they are getting now,” she wrote.
Kalinowski worked at Apple on Mac hardware, then moved to Meta, becoming one of the key engineers at Reality Labs. In 2024 she joined OpenAI as head of robotics and consumer hardware.
In her statement she stressed that the decision was “about principle, not people,” expressing “deep respect” for OpenAI CEO Sam Altman and the startup’s team.
“In my view, the announcement was made hastily, without defining limits. These are first and foremost governance questions. They are too important to rush into deals,” Kalinowski added.
An OpenAI representative stressed that the agreement with the Pentagon “creates a working pathway for the responsible use of AI for national security, while clearly drawing red lines: no domestic surveillance and no autonomous weapons.”
“We understand that people have strong beliefs on these issues, and we will continue to discuss them with employees, government and the public around the world,” he added.
The OpenAI–Pentagon agreement was concluded in late February, shortly after talks between Anthropic and the Department of Defense stalled.
Soon after, Altman acknowledged that the deal was “rushed” and, in reputational terms for the startup, “does not look very good.”
The episode affected the popularity of ChatGPT and Claude. Uninstalls of the former rose 295%, while the latter climbed to the top of the App Store rankings.
Downloads of Claude and ChatGPT mobile apps in the US. Source: Appfigures.The government’s clash with Anthropic
In July 2025 the US Department of Defense signed contracts worth up to $200m with Anthropic, Google, OpenAI and xAI to develop AI solutions for security.
In January 2026 the WSJ reported the agreement with Anthropic was at risk of being terminated. The dispute arose over the startup’s strict ethics policy. The rules prohibit using the Claude model for mass surveillance and autonomous lethal operations.
In February 2026 the US Army deployed Claude in an operation to capture Venezuela’s president Nicolás Maduro, after which the conflict escalated.
The company’s CEO, Dario Amodei, met with defense secretary Pete Hegseth to discuss the situation. The department issued an ultimatum: Anthropic had to accept the government’s terms by 27 February.
The startup’s CEO, Dario Amodei, said he would prefer not to work with the Pentagon than agree to uses of his technologies that could “undermine rather than protect democratic values.”
In late February US president Donald Trump ordered all federal agencies to cease using Anthropic’s technologies entirely within six months.
He called the startup “a company that has spun out of control” run by people who “have no idea what the real world is.”
The Pentagon officially designated Anthropic a “supply chain risk” — putting the firm’s business with other government contractors in jeopardy. Microsoft and Google said they would continue providing products to the company.
Heading to court
Amodei said his company will challenge in court the Department of Defense’s decision to label the firm a supply-chain risk. It is “legally unfounded,” the entrepreneur believes.
The statement says the Pentagon’s decision does not concern the overwhelming majority of Claude’s customers.
According to Amodei, the company held productive talks with the Defense Department for several days, but they fell apart after the leak of an internal memo. In it the CEO described OpenAI’s relationship with the department as “more of a security performance than real risk mitigation.”
Amodei apologized for the leak, saying the company unintentionally disclosed the memorandum’s contents and did not instruct anyone to do so.
“It is not in our interest to make matters worse,” he wrote.
The startup head stressed that the document was written within hours of a series of announcements that Anthropic would be removed from federal systems. He apologized for the tone, calling that day “a difficult one for the company.”
Amodei added that the memorandum is “an outdated assessment” and does not reflect “a careful and considered view.” According to him, the firm’s top priority is to ensure that American soldiers and national-security experts have access to vital tools amid ongoing major combat operations.
Anthropic will continue to provide the Department of Defense with its models “at cost” for as long as necessary.
Former White House artificial-intelligence adviser Dean Ball noted that it will be hard for the company to challenge the Pentagon’s decision.
“Courts are quite reluctant to second-guess government decisions about what is or is not a national-security matter. To do so you need to clear a very high bar. But it is not impossible,” the expert noted.
In February Anthropic softened a key safety policy, explaining it was necessary to remain competitive.