A clause buried in OpenAI’s defense contract is raising serious questions about AI-powered surveillance of American citizens. MIT Tech Review reports that while OpenAI amended its agreement with the Department of Defense to explicitly prohibit domestic surveillance of U.S. persons, legal experts say the language may be largely toothless, as the underlying contract still allows the Pentagon to use the AI system for all “lawful purposes.”
And lawful, it turns out, covers a lot of ground.
The Law Is Stuck in the 1980s
The core problem isn’t the OpenAI contract. It’s that America’s surveillance legal framework is decades out of date. The Fourth Amendment was drafted when “searching” meant physically entering someone’s home. The Foreign Intelligence Surveillance Act of 1978 and the Electronic Communications Privacy Act of 1986 targeted phone wiretaps and email interception. Most of these laws, as MIT Tech Review notes, landed on the books before the internet became a mass phenomenon.
We were not generating terabytes of behavioral data back then. The government did not have tools sophisticated enough to make sense of it. Both of those things are now true.
“The law has not caught up with technological reality,” Alan Rozenshtein, a law professor cited by MIT Tech Review, states plainly.
What AI Actually Makes Possible
Here’s where it gets technically significant. Traditional surveillance required collecting individually sensitive data, which triggered legal protections. AI breaks that logic entirely.
As Rozenshtein explains, “AI can take a lot of information, none of which is by itself sensitive, and therefore none of which by itself is regulated, and it can give the government a lot of powers that the government didn’t have before.”
In practice, that means:
- Aggregation at scale: Combining public social media posts, location pings, purchase histories, and browsing patterns to build detailed profiles
- Pattern inference: Identifying behavioral anomalies or associations without any single data point crossing a legal threshold
- Mass processing: Doing this across millions of people simultaneously, not just targeted suspects
As long as each piece of data was collected lawfully, the government faces few legal barriers to feeding it into AI systems. The problem isn’t the collection. It’s what AI can infer from it.
The Contract Clause That Doesn’t Hold
OpenAI’s amendment states its AI system “shall not be intentionally used for domestic surveillance of U.S. persons and nationals.” That sounds protective. But Jessica Tillipman, a law professor at George Washington University Law School, tells MIT Tech Review it may not change much in practice.
“OpenAI can say whatever it wants in its agreement… but the Pentagon’s gonna use the tech for what it perceives to be lawful,” she says. “Most of the time, companies are not going to be able to stop the Pentagon from doing anything.”
Loren Voss, a former military intelligence officer, acknowledges that legitimate use cases do exist. Counterintelligence missions targeting Americans working for foreign governments or plotting international terrorism are legally authorized. But he also concedes: “This kind of collection does make people nervous.”
The concern is what happens when targeted collection creeps outward, enabled by AI tools that make broader analysis cheap and fast.
What This Means for the AI Industry
This story isn’t just about the Pentagon. It signals a structural tension that every enterprise AI vendor selling to government will face: contractual safeguards written by companies carry less weight than the legal authorities granted to government buyers.
For AI practitioners and businesses operating in the federal space, a few things follow directly:
- Contract language is not a compliance strategy. Legal liability flows from how systems are used, not what sales agreements say.
- Lawful ≠ ethical. AI systems that pass legal review can still enable surveillance that erodes civil liberties. Companies need clearer internal use-case boundaries, not just amended contract riders.
- Regulatory pressure is building. The gap between what AI enables and what the law governs will attract legislative attention. Vendors with opaque government use-case policies will face more scrutiny, not less.
The deeper issue MIT Tech Review surfaces is structural: the rules governing surveillance were written for a world where data was scarce and analysis was slow. AI has reversed both conditions. Until the law catches up, the question of whether the Pentagon is allowed to surveil Americans with AI may have a simpler answer than civil liberties advocates would prefer: largely, yes.