Congress Asked AI Companies If They'd Help Spy on Americans

The United States Capitol dome illuminated at dusk against a dark blue sky — Photo via Unsplash

TL;DR: On March 4, 2026, Senator Ron Wyden sent letters to four AI companies (Anthropic, Google, OpenAI, and xAI) asking a direct question: would you let the government use your AI to analyze bulk commercial data on Americans? Only Anthropic and Google responded. Anthropic said it bans bulk domestic surveillance but disclosed it has granted exceptions to "a small number of national-security customers" for foreign intelligence analysis, even when that analysis "includes incidentally collected U.S.-person information." Google emphasized privacy protections. OpenAI and xAI (the latter owned by Elon Musk) said nothing. Meanwhile, federal agencies are spending billions on surveillance tools and AI capabilities, and FISA Section 702 gives them warrantless access to Americans' communications. The question isn't whether AI will supercharge government surveillance. It already has.

The Question No One Wanted to Answer

Wyden's letter was simple. He asked the CEOs of America's four leading AI companies whether their products could be used by government customers to analyze commercial data (location information, web browsing histories, advertising records): the kind of data the government already buys from data brokers without warrants.

The letter went to Dario Amodei at Anthropic, Sundar Pichai at Google, Sam Altman at OpenAI, and Elon Musk at xAI. It asked whether their companies would "allow government customers to use their products to engage in practices that may be technically legal, but that violate privacy" ^[1].

Wyden framed it as a yes-or-no question about a specific pipeline: federal agencies buy bulk personal data from brokers, then feed it into AI systems that can identify patterns, predict behavior, and build profiles at a speed and scale no human analyst could match. Would the companies help build that pipeline?

Two answered. Two didn't.

What Anthropic Actually Said

Anthropic's response, through its head of North America government affairs Brian Peters, was the most detailed, and the most revealing.

The company said its usage policy bans "unauthorized surveillance or tracking of individuals" and bars "analysis of the product of bulk domestic collection." That sounds reassuring. But keep reading.

Peters disclosed that Anthropic had granted an exception "to a small number of national-security customers, permitting the use of our models for foreign intelligence analysis in accordance with applicable law." And here's the clause that matters: that analysis can include data that "includes incidentally collected U.S.-person information" ^[2][3].

If you've followed the FISA Section 702 debate, you know what "incidentally collected U.S.-person information" means. It's the legal term for Americans' communications that get swept up when the government targets foreigners. The FBI searched this data over 7,000 times in 2025 alone, a 34% increase from the prior year, without warrants ^[4].

CEO Dario Amodei put the stakes in plain language: "Powerful AI makes it possible to assemble this scattered, individually innocuous data into a comprehensive picture of any person's life, automatically and at massive scale" ^[1].

He's right. And his company just told Congress it's letting national security customers do exactly that, with carve-outs.

Google's Reply: Privacy Language, Government Contracts

Google responded too, emphasizing its commitment to protecting privacy and constitutional rights. But Google's answer carries its own baggage.

DHS has issued "hundreds" of subpoenas to Google, Reddit, Discord, and Meta for user data ^[5]. Google holds some of the most comprehensive personal data collections on the planet: search histories, location data, email contents, device information. When the government asks for it, Google has a track record of compliance.

Saying you care about privacy while sitting on the world's largest advertising surveillance infrastructure requires a certain comfort with contradiction. Google's AI tools are already embedded in government operations. The question isn't whether Google will participate in government surveillance. It's whether there are meaningful limits on how.

OpenAI and xAI Said Nothing. That's an Answer.

OpenAI, the company that announced a Department of Defense deal on February 28, didn't respond to Wyden's letter. Neither did xAI, Elon Musk's AI company.

OpenAI's DoD deal initially described banning use for "mass domestic surveillance." Days later, after public criticism, OpenAI amended the agreement to prohibit "domestic surveillance of U.S. persons and nationals." The shift from "mass" to "any" matters, but the fact that it took public pressure to get there matters more.

OpenAI chose not to tell a sitting U.S. senator whether its technology can be used to spy on Americans through bulk data analysis. That's not a non-answer. It's a policy position.

xAI's silence is less surprising but equally telling. Musk has positioned himself as a government ally. His companies already hold federal contracts. When a senator asks whether your AI will be used for domestic surveillance and you don't respond, the silence speaks.

The Pipeline Already Exists

This isn't hypothetical. The surveillance pipeline Wyden described, data brokers selling personal information to agencies that feed it into AI systems, is already operational.

Here's what federal agencies are spending:

DHS signed a $1 billion contract with Palantir in 2026 to deploy AI-powered analytics across all DHS components, including ICE and CBP ^[6]
ICE's surveillance budget hit $28.7 billion, ten times what the agency spent on surveillance tools over the previous 13 years combined
DHS awarded Penlink PLX a $2.9 million no-bid contract for real-time location tracking through ad exchange data, with a potential value of $8.3 million ^[7]
The FBI increased warrantless searches of Americans' Section 702 data by 34% in 2025, to over 7,000 queries

Every one of these programs can be enhanced by AI. Pattern recognition across location databases. Natural language processing on intercepted communications. Behavioral prediction from browsing histories. The tools exist. The data exists. The legal authority, through Section 702 and the data broker loophole, exists. The only question is which AI companies will say no.

FISA 702 Makes It Legal. AI Makes It Easy.

Section 702 of the Foreign Intelligence Surveillance Act was designed to target foreign communications. But it sweeps up Americans' data as a side effect, and the FBI can search that data without a warrant.

Now add AI. Patrick Eddington at the Cato Institute put it directly: searching massive surveillance databases with AI can be done "in a matter of seconds." And if the database contains errors (which it does), "that's going to get carried all the way through to the end" ^[2].

Wyden warned on the Senate floor: "The government could use AI to process huge amounts of disparate data to identify Americans for warrantless searches" ^[8].

The 45-day FISA extension expires June 12. Congress returns from recess Monday. They have 33 days to decide whether AI-powered surveillance gets the same legal framework as manual searches, or whether the current system (collect everything, search anything, warrant optional) gets a three-year renewal with AI on top.

Wyden's bipartisan Government Surveillance Reform Act would require warrants for accessing location data, web browsing records, and establish AI guardrails in intelligence agencies. Whether it gets a vote depends on what happens in the next four weeks ^[9].

What This Means for You

If AI companies won't draw clear lines on government surveillance, the lines don't exist. Here's what you can do:

Use a VPN and privacy-focused browser. Your browsing data is sold to data brokers who sell it to the government. Firefox with uBlock Origin, or Brave, reduces your advertising footprint. A reputable VPN masks your IP from broker-tracked ad requests
Opt out of data brokers. Services like DataPurge help automate removal requests. The fewer databases you're in, the less the government can buy without a warrant
Contact your senators about FISA reform. The Government Surveillance Reform Act (S. 4082) would require warrants for Americans' data and add AI guardrails. Congress is back Monday. The deadline is June 12. Calls matter now
Track which AI companies hold government contracts. OpenAI's DoD deal, Google's cloud contracts, Palantir's DHS work: these are public records. Know which tools your data might be fed into

The Bottom Line

Wyden asked a simple question. Two companies gave qualified answers. Two didn't answer at all. And the company with the most explicit "no surveillance" policy (Anthropic) still carves out exceptions for national security customers analyzing data that includes Americans' information.

Meanwhile, the government is spending billions on AI-powered surveillance infrastructure, buying bulk personal data without warrants, and searching Americans' communications at record rates. FISA Section 702 gives them the legal cover. AI gives them the capability. And the companies building the most powerful AI systems in the world are either drawing fuzzy lines or refusing to draw any at all.

The convergence of AI and government surveillance isn't a future risk. It's the present. The only question left is whether Congress will act before June 12, or hand the surveillance state a three-year blank check with AI built in.

References

Published: May 10, 2026