The announcement of OpenAI’s Pentagon AI deal has arrived at a moment of growing public concern about the role of artificial intelligence in warfare, adding urgency to questions about whether any commercial AI company can truly guarantee its technology will not be used for autonomous killing. The deal comes in the wake of Anthropic’s very public expulsion from government contracts for refusing to permit exactly that.
Anthropic had taken a clear and consistent position: its Claude AI would not be used to power weapons systems capable of making lethal decisions without human oversight, and it would not be used for mass surveillance of civilian populations. The company considered these not edge-case concerns but central questions about AI’s impact on society, and it held to them through months of Pentagon pressure.
The administration’s decision to ban Anthropic from all federal contracts was framed by President Trump as a defense of constitutional principles against corporate interference. But Anthropic — and much of the AI workforce — saw it differently: as an attempt to normalize the use of AI for autonomous weapons and mass surveillance by eliminating companies that refused to participate in those uses.
Sam Altman’s Pentagon deal, announced hours after Anthropic’s expulsion became final, included what he described as contractual protections against both of those uses. He framed the agreement as evidence that principled government engagement is achievable and called for the terms to be standardized across the industry. His words tracked closely with Anthropic’s stated principles, raising the question of why the two companies reached such different outcomes.
The AI workforce’s solidarity with Anthropic — expressed in a letter signed by hundreds of workers from OpenAI and Google — suggested that many inside the industry share Anthropic’s concern about autonomous weapons and mass surveillance. Anthropic’s response to its political exile was composed and resolute: it had offered everything lawful and held firm on only two narrow exceptions, and that position would not change regardless of the government’s response.