Sam Altman had been saying for months that OpenAI was a national security company. Not in the sense that it worked for governments, but in the sense that developing the most advanced AI possible was, in itself, a security issue for the United States against China and other competitors. It was an argument designed to set the stage for what ended up happening on February 28, 2026: OpenAI signed a contract with the Pentagon to deploy its models on the Department of Defense's classified and unclassified networks.
The news was shocking not because no one expected it—it had been brewing for months in the halls of Washington—but because of the exact context in which it occurred: the same day that President Donald Trump announced that all federal government agencies must stop using the artificial intelligence tools of Anthropic, OpenAI's direct competitor. The timing was too perfect to be a coincidence.
What the contract actually allows
According to available public documents, the agreement allows models like ChatGPT to operate in national security environments for specific tasks: logistics optimization and management of military supply chains, intelligence analysis and processing of classified information, and support in operational planning and strategic simulations.
OpenAI established two public limits. First: prohibition of the use of its technology for mass domestic surveillance of American citizens. Second: prohibition of the use in lethal autonomous weapons systems without direct human supervision. Both limits sound reasonable. The problem is what Altman later admitted, in internal statements that were reported to the media: that operational decisions about the deployment of AI remain exclusively in the hands of US military authorities. That OpenAI has no control over the Pentagon's use of its technology once delivered.
Altman admitted internally that OpenAI has no control over what the Pentagon does with its AI once the contract is signed. The limits he publicly announced are, in practice, unilateral promises with no verification mechanism.
Caitlin Kalinowski: the resignation that no one expected
The most eloquent response to the contract announcement did not come in the form of an opinion article or an institutional statement. It came in the form of resignation. Caitlin Kalinowski, head of OpenAI's robotics team — the area that defines the next big frontier of AI, physical systems that operate in the real world — left her position days after the announcement, publicly citing deep concerns about military use of the technology she had helped build.
Kalinowski's departure was not that of someone who disagrees with a minor strategic shift. It was that of someone who believes that the founding values of his work have been compromised in ways that he cannot accept professionally. He leads the area of robotics—systems that in the military context can mean autonomous drones, unmanned vehicles, or combat robots—and his resignation is precisely in that context a signal that is impossible to ignore.
Chronology of the OpenAI–Pentagon contract
- January 2026: Defense Department Hegseth requires AI contracts to grant use for "any lawful purpose" without vendor restrictions
- February 28, 2026: Trump bans the use of Anthropic in federal agencies · OpenAI signs contract with the Pentagon on the same day
- March 3, 2026: Altman announces modifications to the agreement after internal and external pressure
- March 7, 2026: Caitlin Kalinowski, head of robotics, resigns over "concerns about surveillance and autonomous weapons"
- May 1, 2026: Pentagon makes official contracts with 8 companies for classified military networks (IL6/IL7)
The problem of guardrails that cannot be verified
When Altman announced the modifications to the contract—the ban on mass surveillance, the requirement for human oversight in the use of force—he did so in response to the internal and external pressure that followed the initial announcement. The declaration of intent is public. But the verification mechanism does not exist. There is no independent body that can audit whether the Pentagon respects those limits. There are no automatic termination clauses if violated. There is no transparency about the specific projects in which the technology is used.
This is not an accusation that the Pentagon will violate the agreements. It is the observation that the agreements, as structured, are unilateral promises by one party—OpenAI—about the behavior of the other party—the Department of Defense—without any real accountability mechanism. In any other context of technological regulation, that would be considered insufficient.
Altman's argument and its problems
Altman's public defense of the contract rests on an argument that has some internal coherence but is also deeply problematic. The argument is: if the United States doesn't develop advanced military AI, China will. And if China does it, it will do it without any of the ethical safeguards that OpenAI wants to implement. Then it is better that OpenAI provides the technology, because at least it has principles.
The problem with this argument is twofold. First: assume that OpenAI can actually maintain those principles once the customer is the Pentagon with authority over the final deployment. Kalinowski's resignation suggests that at least within the company there are people who seriously doubt him. Second: the argument of "if we don't do it, someone else will do it worse" is one of the most historically used reasoning to justify participation in projects that are abstractly recognized as problematic. Just because it is an understandable argument does not mean it is correct.
Relevant precedent: In 2018, thousands of Google employees protested and some resigned when the company signed Project Maven with the Pentagon for drone image analysis. Google did not renew that contract. In 2026, OpenAI went in the opposite direction — and the internal response has been considerably quieter, at least publicly.
What it means for the future of ethical AI
The OpenAI–Pentagon contract closes a cycle that began when the company was founded with the explicit objective of developing AI safely and for the benefit of humanity. It's not that that objective has disappeared from corporate documents. It's just that the concrete decisions being made—first the restructuring into a for-profit company, then the military contract—are effectively redefining what that goal means.
Kalinowski's departure is a reminder that behind AI companies there are people with specific values who at some point may decide that those values are not compatible with the direction the company is taking. Such people do not usually make loud statements. They just leave. And when they leave, they take with them something that money and press releases can't easily buy: moral credibility.
Share this article
