Background on the OpenAI-Pentagon Agreement
OpenAI CEO Sam Altman publicly defended his company’s new Pentagon deal on Saturday, a day after President Donald Trump ordered federal agencies to cut ties with rival Anthropic. Hours after a U.S.-Israel joint strike against Iran, Altman used X to answer questions about the agreement permitting the Department of War (DoW) to deploy OpenAI’s artificial intelligence models on its classified network.
I’d like to answer questions about our work with the DoW and our thinking over the past few days.
Safety Principles in the Agreement
In announcing the agreement late Friday, Altman emphasized that AI safety and wide distribution of benefits form the core of OpenAI’s mission. Two key safety principles prohibit domestic mass surveillance and mandate human responsibility for the use of force, including autonomous weapon systems. The DoW agrees with these principles, reflects them in law and policy, and they are incorporated into the agreement.
Trump’s Directive Against Anthropic
The OpenAI agreement emerged as Trump directed every federal agency to stop using Anthropic technology, with a six-month phase-out period, intensifying disputes over AI in military operations. Secretary of War Pete Hegseth designated Anthropic a supply-chain risk to national security. Anthropic CEO Dario Amodei refused DoW demands to allow its AI for all lawful purposes, citing concerns about mass domestic surveillance and fully autonomous weapons.
Anthropic seemed more focused on specific prohibitions in the contract, rather than citing applicable laws, which we felt comfortable with. He added that Anthropic may have wanted more operational control than we did.
Negotiations and Shift to Classified Work
Altman stated the DoW issued no explicit or implicit threats before the agreement, and Pentagon officials were surprised OpenAI considered classified work. Initially planning only non-classified work, talks accelerated this week into classified applications. OpenAI had rejected prior classified deals that Anthropic accepted, but found the DoW flexible on needs and supportive of its mission.
We thought the DoW clearly needed an AI partner, and doing classified work is clearly much more complex. We have said no to previous deals in classified settings that Anthropic took.
Addressing Criticisms and Future Risks
Altman addressed perceptions of a rushed deal, noting OpenAI acted quickly to de-escalate. He negotiated terms to extend similar offers to all other AI labs, warning the current path endangers Anthropic, healthy competition, and the U.S. While hopeful for resolution, he acknowledged potential supply-chain risk designations for OpenAI. On nationalization concerns, Altman deemed it unlikely but stressed close government-AI company partnerships. The most difficult reconciliation involved non-domestic surveillance, which he accepts pragmatically while respecting the democratic process.
I have accepted that the US military is going to do some amount of surveillance on foreigners, and I know foreign governments try to do it to us, but I still don’t like it.
