Note to staff: Sam Altman says OpenAI seeks a DOD deal, except for use cases like domestic surveillance, and wants to “help de-escalate” DOD-Anthropic fight

Wall Street Journal 2026-02-27 Keach Hagey

Context & Ripple Effects

Altman framed a prospective DOD relationship as compatible with a boundary against domestic surveillance, while also positioning OpenAI as a potential bridge in the Pentagon-Anthropic dispute. That makes the issue not simply government procurement, but the terms on which a frontier-model provider accepts defense work.

The stance was followed by reports that OpenAI sought explicit contractual limits on domestic surveillance and that the DOD would allow OpenAI to build its own safety stack. Later internal discussion also exposed the harder limit of provider control: OpenAI said it does not make the military's operational decisions.

First-order effects

OpenAI’s DOD negotiations become contingent on defined use-case restrictions, putting domestic surveillance at the center of the company’s public contracting posture.
The DOD-Anthropic disagreement gains a second commercial and reputational dimension as Altman offers OpenAI as a participant in de-escalation rather than only a prospective supplier.

Second-order effects

Anthropic and other AI vendors face pressure to articulate whether their defense-access policies rely on categorical prohibitions, contractual safeguards, or provider-controlled safety systems.
Contract language and technical enforcement become competitive differentiators: later reporting on limits on access by agencies such as the NSA shows that customer eligibility and downstream use can require separate negotiation.

Third-order effects

If these arrangements proliferate, defense AI procurement may shift from broad model access toward negotiated governance packages combining permitted users, prohibited applications, and vendor safety controls.
The unresolved question is whether contractual restrictions can remain meaningful once systems are deployed in military workflows where providers acknowledge they lack operational authority.

The trend: Frontier AI labs are becoming state-compatible suppliers by negotiating use boundaries and governance control alongside model access.

Discussion

@scaling01 @scaling01 on x
he wants to be part of this so bad he knows this was an insane relative aura loss for labs not named Anthropic
@hadas_gold Hadas Gold on x
Sam Altman on Pentagon/Anthropic fight just now on CNBC: “The government the Pentagon needs AI models. They need AI partners. This is clear and I think Anthropic and others have said they understand that as well. I don't personally think the Pentagon should be threatening DPA
@mikeisaac Rat King on x
OAI waits to see where wind blows in showdown with pentagon. anthropic doesn't flinch, get lots of cred among developer community yesterday we see open letter from rank and file at OAI/goog support anthropic. now, OAI writes this designed to leak to claim some goodwill funny
@andrewcurran_ Andrew Curran on x
Sam Altman in a note to staff last night: 'We are going to see if there is a deal with the DoW that allows our models to be deployed in classified environments and that fits with our principles. We would ask for the contract to cover any use except those which are unlawful or [im…
@isolyth.dev @isolyth.dev on bluesky
it appears that OpenAI is adopting the same restrictions as Anthropic, or is looking to, though their language is a fair bit softer than Anthropic's here
@hshaban Hamza Shaban on x
OpenAI's Sam Altman is pursuing a deal with the Pentagon “that allows our models to be deployed in classified environments and that fits with our principles,” he wrote in a note to staff Thursday evening, the WSJ reports [image]

Chronicles