OpenAI and the DOD agree to add more surveillance protections to their AI deal; sources say Sam Altman approached the DOD's Emil Michael to rework the contract
OpenAI and the Pentagon have agreed to strengthen their recently agreed contract, following widespread backlash …
AxiosMaria Curi
Context & Ripple Effects
OpenAI’s move into the Defense Department’s classified network was already framed as a bid to extend comparable access across the AI sector. The subsequent disclosure that the original terms left room for bulk data analysis triggered a visible dispute over how that access would be governed.
OpenAI and the DOD must incorporate stronger surveillance protections into the contract governing the reported deployment, narrowing the permitted use cases relative to the disputed prior terms.
OpenAI gains a clearer public basis for defending the deal, while the DOD accepts more explicit limits on how the AI system may be used for domestic surveillance.
Second-order effects
The revision creates a concrete contractual benchmark for other AI providers pursuing DOD access; OpenAI had already called for its classified-network terms to be extended to all AI companies in its proposed industry-wide approach.
Future negotiations will likely focus more directly on operational guardrails and auditability, rather than treating compliance with existing law as a sufficient answer to surveillance concerns.
Third-order effects
If such clauses become standard, defense adoption of frontier models may be shaped increasingly by vendor-specific use restrictions as well as government procurement rules.
The episode reinforces that strategic AI partnerships require continuing legitimacy management: public backlash can reopen terms even after a high-profile government deployment is agreed.
The trend: Dual-use AI procurement is evolving from model-access agreements into negotiated governance regimes that determine acceptable state uses.
In the last few days, OpenAI and its executives have claimed that its DoW deal prevents its models being used for mass domestic surveillance. As I write in a lengthy explainer for @ReadTransformer today, that appears to be misleading at best. [image]
Great to see OpenAI clarify and explicitly re-write the terms of their government deal! They explicitly ban their AI from being used for mass surveillance and by intelligence service Good for them 👏👏
Strategic miscalculations from OpenAI starting to add up: 1. Taking on Google and Apple and Meta at the same time 2. Public crash out on prominent shareholder's podcast 3. DoW contract
For people just getting to know the Department of War's Emil Michael there are two big things you need to know about his time as the no 2 to Kalanick at Uber. 1/ he was the architect of Uber's war chest. In this respect he was the sam altman before sam altman, scouring the
@ShakeelHashim @ReadTransformer New language reinforces that domestic surveillance is disallowed under this agreement, including involving commercially acquired information. See @sama 's post.
Claude is so much better than ChatGPT. Jesus Christ lol. I didn't join it because of some hippie dippy nonsense about pentagon contracts. I joined it for the normal reason (people on twitter said it was good). And damn howdy it sure is better than ChatGPT
2/ he was notorious as the exemplar of Uber's aggressive culture. Someone under him was accused of accessing a rape victim's confidential medical records (Uber settled) and he ultimately resigned around the same time as Kalanick due to the company's myriad scandals.
Sam Altman just said he has tweaked his deal with the Pentagon and wrote a list of things he says are part of it, and he chose to include this really great line: — “One thing I think I did wrong: we shouldn't have rushed to get this out on Friday.”
“The problem with telling everyone what they want to hear is that eventually reality catches up with you. The people who will live under AI-powered surveillance, and the people in the flight path of AI-assisted drone swarms...” — www.platformer.news/openai-penta...