Anthropic refused to accept the new conditions set by the Pentagon for the continuation of their cooperation, firmly insisting on its refusal to develop mass surveillance technologies and fully autonomous lethal weapons. This decision comes after months of negotiations and an ultimatum issued by Defense Secretary Pete Hegseth.
See also: Anthropic: Chinese AI companies are trying to copy Claude
Anthropic's stance sets an important precedent for the AI industry, as it shows that some companies are willing to sacrifice lucrative contracts to maintain their ethical standards.
The initial agreement between Anthropic and the Department of Defense was signed in July 2025, for a value of up to $200 million over two years. The agreement was part of a broader Pentagon strategy to integrate advanced AI technologies into its defense capabilities, particularly in the areas of intelligence analysis and operational support. The agreement included the development of prototype AI systems that could process massive amounts of data in classified environments. Similar contracts had been signed with OpenAI, Google, and xAI, creating a competitive landscape in the defense artificial intelligence sector.
Dario Amodei, CEO of Anthropic, stressed that the company remains committed to supporting national defense, but not at the cost of democratic values. In a statement, Amodei reiterated the company’s stance against mass surveillance and the use of fully autonomous weapons, underscoring the importance of ethical considerations in the development of AI technologies. This position reflects a broader philosophy at the company that prioritizes safety and aligning AI systems with human values, even when it means losing significant revenue.
See also: Anthropic: Claude offers more free features
Negotiations between Anthropic and the Pentagon have intensified in recent months, with the company insisting on safeguards against mass surveillance of American citizens and fully autonomous lethal weapons. The disagreements centered on specific clauses that would allow the Pentagon to use Claude technology for applications that Anthropic considered ethically problematic. In late February 2026, Defense Secretary Pete Hegseth issued an ultimatum with a deadline of the end of the week, sending a “final and best offer” that Anthropic rejected without hesitation.
According to sources familiar with the negotiations, Anthropic believed the new terms contained insufficient protections and “loopholes” in the wording that would allow the Pentagon to bypass ethical safeguards. The company expressed concerns that apparent concessions in the agreement’s language could cover discretionary circumventions on the military’s part.
Anthropic had already developed custom Claude Gov models on AWS infrastructure for federal agencies, integrating its systems with partners like Palantir to process “massive amounts of complex data” across distributed networks. These systems were designed to support analytics, risk prediction, and operational support, but without violating the company’s core principles.
See also: Anthropic says Claude will remain ad-free
Thiyagu Ramasamy, head of public sector at Anthropic, had described the initial agreement as promoting “responsible AI in defense,” emphasizing the importance of balancing efficiency and ethics.
