This week’s contrasting fortunes of OpenAI and Anthropic have produced two distinct lessons for the AI industry about engaging Washington. The first is that engagement is possible and commercially rewarding. The second is that resistance, even principled resistance, carries a severe commercial cost in the current political environment. How companies weigh these lessons will shape the industry’s relationship with government for years to come.
Anthropic’s resistance was principled and specific. The company refused to allow two particular uses of its Claude AI system — autonomous weapons and mass surveillance — while offering broad support for every other lawful military application. This was not general obstruction but a carefully limited ethical position that the company had articulated publicly and consistently since the beginning of its Pentagon negotiations.
The cost of maintaining those principles was immediate and severe. President Trump’s ban on all federal use of Anthropic technology, announced in characteristically combative terms on Truth Social, ended the company’s government relationships overnight and demonstrated the administration’s willingness to use political power to discipline AI companies that resist its demands for unrestricted access.
OpenAI’s engagement produced a different outcome. Sam Altman announced a Pentagon deal with assurances that the agreement includes the very protections Anthropic had insisted on — protections against mass surveillance and autonomous weapons use. He simultaneously raised $110 billion in new funding, demonstrating that commercial success and government partnership are not mutually exclusive even in a politically charged and volatile environment.
The industry is left to weigh these lessons against each other. Hundreds of workers across OpenAI and Google signed letters backing Anthropic, suggesting that within the companies themselves, sympathy lies with resistance rather than engagement. Anthropic’s own statement — declaring its principles non-negotiable and its restrictions never once harmful to a legitimate mission — frames the company’s exile not as a mistake but as an act of integrity. Whether the industry ultimately rewards integrity or pragmatism will determine what kind of AI it builds and for whom in the years ahead.