The events of this past week have made brutally clear the impossible choices now facing AI companies as the US government asserts dominance over how artificial intelligence can be deployed. OpenAI has chosen a deal; Anthropic has chosen its principles. The industry is left to wonder whether any company can truly have both.
Anthropic’s refusal to allow its Claude AI to be used for autonomous weapons or mass surveillance reflected the company’s foundational commitment to safety-focused AI development. These were not minor commercial quibbles but fundamental positions on the role of human oversight in life-or-death decisions and the protection of civil liberties.
The Trump administration’s response — a complete, immediate, and very public ban on Anthropic’s government use — demonstrated that the current political environment has little tolerance for AI companies that attempt to govern how their technology is used in national security contexts. The message was clear: comply or be cut off.
Sam Altman’s announcement of a Pentagon deal with built-in ethical commitments represents a different approach — one that attempts to maintain principle while remaining inside the tent. His call for the government to standardize these terms across the industry suggests he sees OpenAI’s deal not as a departure from Anthropic’s values but as a more effective way of advancing them.
The hundreds of workers who publicly backed Anthropic and warned against government pressure suggest the AI workforce sees things differently. Anthropic itself has not flinched, insisting that its restrictions are absolute and that they have never, as far as the company can determine, blocked a single legitimate national security operation.