This week, in a dramatic escalation in how artificial intelligence is integrated into national security and who controls it, Anthropic was officially blacklisted by the Trump administration and OpenAI rushed in to win new defense contracts.
On Friday night, the Pentagon designated Anthropic as a supply chain risk, barring defense contractors from using the technology after a transition period. This comes just hours after President Donald Trump directed federal agencies to stop using Anthropic’s AI tools because the company refused to agree to the military’s proposed use of the Claude model.
Anthropic CEO Dario Amodei said he could not “in good conscience” allow the technology to be used for domestic mass surveillance or to independently direct autonomous weapons, two use cases he said violated the company’s ethical guardrails.
Amid the standoff, Anthropic’s rival OpenAI announced it had reached an agreement with the Department of Defense to deploy its proprietary AI models in sensitive environments.
The dispute has evolved into a broader conflict between the Pentagon and the civilian AI sector, and is not just about military contracts, but about who ultimately sets the terms for how these powerful systems are used.
Conflict of principles and contracts
Anthropic argued that restrictions on the monitoring of its systems and the use of autonomous weapons systems were not adequately reflected in the government’s draft contract language and were not enforceable.
Defense officials responded that they needed to be able to deploy Claude for “lawful uses,” even though mass domestic surveillance is illegal under several statutes. This term gives the military wide discretion.
Dean Ball, a senior fellow at the American Innovation Foundation, described the conflict as “uncharted territory” rooted in competing principles. Anthropic has argued for contractual limits on how its technology can be used, and the Pentagon has said defense policy should take precedence over corporate priorities.
“This is a matter of principle on both sides,” Ball told Business Insider.
Shortly after Anthropic was declared a supply chain risk, OpenAI announced in a blog post that it had agreed to a framework with the Department of Defense that explicitly included guardrails similar to some of what Anthropic was seeking.
OpenAI has outlined three key “red lines” to guide its collaboration with the Department.
- The technology will not be used for large-scale domestic surveillance.
- Directing autonomous weapons is useless.
- Also, other high-stakes automatic decisions, such as “social credit” systems, are useless.
OpenAI says its agreement is structured to protect these restrictions with layered security measures, and any use involving autonomous weapons or surveillance must comply with existing legislation and ministerial directives.
OpenAI CEO Sam Altman said in a series of “Ask-Me-Anything” posts that the company is prepared for potential disputes over the legality of certain government requests. But he said OpenAI would not agree to allow governments to use its technology for domestic mass surveillance.
“I fear a world where AI companies act as if they have more power than governments,” Altman added. “A world in which our government decides that mass domestic surveillance is OK would also be scary.”
Potential “existential” interests
In this context, legal scholars say the government’s threat to enact the Defense Production Act and designate artificial enterprises as supply chain risks is unusual.
Eric Chaffee, a business law professor at Case Western Reserve University, called the move a “gamble,” noting that the recent Supreme Court decision overturning President Trump’s tariff policies sets back broad executive action without clear legal support.
And while national security officials are usually respected in court, any case would be played out in a legal limbo.
“At this point, the government ends up contracting out a lot of the work to private entities,” Chaffee said. “As a result, there will be tensions arising from negative attitudes toward state-owned enterprises, and ultimately it will be a complex process to understand how private companies engage with governments in these situations.”
For the Department of Defense, removing Anthropic from the military AI pipeline could cause confusion. George Pollack, policy analyst at Signum Global Advisors, said Claude is deeply embedded in defense planning and readiness systems, and moving from there risks creating operational friction.
he They also argued that sidelining major U.S. AI companies contradicts the stated goal of maintaining U.S. leadership in technology.
For Anthropic, the bet is about “survival,” Ball warned. He said the dispute could send a chilling message to entrepreneurs about the risks of doing business with the federal government, if companies could face penalties for insisting on ethical guardrails.
OpenAI’s agreement with the department appears to have been aimed, in part, at breaking an impasse involving negotiations with Anthropic. OpenAI said in a blog post that it called for the same terms to be made available to all AI labs and called on the government to resolve the dispute with Anthropic.
Representatives for OpenAI and Anthropic did not immediately respond to requests for comment from Business Insider. It was not immediately clear whether Anthropic or other large AI companies were offered similar terms to those OpenAI said it had agreed to.
Ultimately, how this dispute is resolved could affect not just the fate of one company, but also the balance of power between the U.S. government and private AI developers in setting the rules of engagement for next-generation technologies.
