The Government Can Already Buy Your Location Data Without a Warrant. Now It Wants AI Too.

In the last week of February 2026, something extraordinary happened. Most people missed it entirely because it unfolded not in congressional hearings or courtrooms, but in a flurry of social media posts and corporate blog entries.

Anthropic, the company behind the AI assistant Claude, had been negotiating a contract with the U.S. Pentagon to allow the military to use its technology. Those negotiations collapsed. Publicly. Dramatically. And in a way that should concern every American who carries a smartphone.

Here’s what happened: Anthropic had agreed to let the Pentagon use Claude, but with two firm exceptions written into the contract. Its AI would never be used for mass domestic surveillance of American citizens, and it would never power fully autonomous weapons, so-called “murder bots” that could make kill decisions without human authorization.

The Pentagon wanted those exceptions removed.

Anthropic CEO Dario Amodei published a statement saying the company would not budge, regardless of consequences. In a tech world that has spent the past year falling over itself to please the current administration, the line that stood out was this: “These threats do not change our position. We cannot in good conscience accede to their request.”

Within 24 hours, President Trump posted on Truth Social calling Anthropic a “radical left woke company” and ordering all federal agencies to stop using its technology. An hour later, Defense Secretary Pete Hegseth went further, posting on X that he was designating Anthropic a supply chain risk. That designation is normally reserved for adversarial foreign entities like Chinese chip manufacturers, and no American company of Anthropic’s size had ever faced it, as far as anyone can determine. That threat became reality this week: the Department of Defense formally notified Anthropic’s leadership that the designation is official and effective immediately, requiring every defense vendor and contractor in the country to certify they are not using Claude in their work with the Pentagon. Anthropic says it will fight the designation in court.

Then, in the immediate chaos of that first night, something else happened that made an already confusing situation nearly impossible to parse. OpenAI CEO Sam Altman announced they had signed a deal with the Pentagon, claiming they had secured the exact same protections Anthropic had just been punished for demanding.

The Three Words That Explain Everything

To understand why this matters beyond the drama, you need to understand three words: “all lawful use.”

That was the Pentagon’s demand. Strip out Anthropic’s specific exceptions, and replace them with a simple standard: the military can use this AI for anything that is legal.

On the surface, that sounds reasonable. Of course the government shouldn’t do anything illegal. But here is the problem worth understanding: the United States has no national privacy law, and no meaningful regulation governing how AI can be used. That means the category of “lawful” is very wide. Wider than most people realize.

Anthropic’s argument was not that the Pentagon intended to break the law. Their argument was that what currently passes for legal in this country already includes things that many Americans would recognize as surveillance. Handing over a powerful AI system under an “all lawful use” standard would allow it to be used in ways that have no legal obstacle whatsoever.

The evidence for that concern is already hiding in plain sight.

Your Phone Is Already Feeding This System

Here’s what most people don’t know: the government doesn’t need to hack your phone to track you. It just needs to buy the data.

Right now, dozens of companies called data brokers collect precise information about where you go, what you buy, what you search for, and what you post. That data is gathered largely from the apps on your smartphone. Your weather app. Your navigation app. Your coupon app. When you give those apps location access to get directions or a forecast, many of them quietly share that data with third parties who package it and sell it to anyone willing to pay.

Including federal agencies.

The ACLU has been litigating this for years. Their investigations found that ICE purchased access to a surveillance system specifically designed to monitor a city neighborhood for mobile phones, track movements over time, and follow people from their workplaces to their homes. The Department of Homeland Security spent over two million dollars on contracts with a single location data company called Venntel, which provided “geofence” capabilities. Geofencing is the ability to identify every device present in a specific area at a specific time. No warrant required for any of it.

The intelligence community is now moving to centralize all of this. The Office of the Director of National Intelligence is building a portal to streamline the purchase and use of this kind of commercially available data across multiple spy agencies. As The Intercept reported in May 2025, this is data that would previously have required a court order to obtain.

The Brennan Center for Justice, a nonpartisan law and policy institute at NYU, calls it “buying their way around the Fourth Amendment.” The Fourth Amendment, for those who need a reminder, is the part of the Constitution that protects Americans from unreasonable searches. It was written before weather apps existed.

“Nobody expects that by carrying a phone, they are somehow consenting to let the government make a record of their every move.” — American Civil Liberties Union

This is what Anthropic was actually worried about. Not some hypothetical future. The infrastructure for mass surveillance of ordinary Americans is already built. What AI adds to this picture isn’t the capability to collect the data, because that already exists. What AI adds is the ability to analyze all of it, instantly, at a scale no human team could ever manage.

An AI system with “all lawful use” permission, fed into this existing data pipeline, could cross-reference your location history, your purchase records, your social media activity, and your associations in real time, across millions of people simultaneously. All legally. All without a warrant. All right now.

This Isn’t Hypothetical. It’s Already Happening.

If that still feels abstract, consider what is already underway.

A March 2026 NPR investigation found that ICE has built a sweeping surveillance network that extends beyond undocumented immigrants to include legal observers, activists, and people posting criticism of the agency online. People are being tracked, confronted, and in some cases detained based on their digital activity.

Is that domestic surveillance? The Pentagon would say no. It’s immigration enforcement, and it’s legal. But Anthropic’s position was that if you hand a powerful AI system an “all lawful use” permission slip in a country where that kind of monitoring is legal, you have handed over a tool that can be directed at anyone posting the wrong opinion at the wrong time.

Surveillance infrastructure doesn’t stay pointed in one direction.

The Other Side of the Argument

Before going further, it’s worth being fair to the Pentagon’s position, because it isn’t unreasonable on its face.

The military’s core argument is that a private company should not be able to dictate the operational terms of a sovereign nation’s defense. No other contractor, whether a weapons manufacturer, a logistics firm, or a satellite provider, gets to carve out specific use cases and say “not for that.” The Pentagon’s view is that existing laws, congressional oversight, and military command structures are the appropriate checks on how AI gets used, not the preferences of a tech company’s leadership. Emil Michael, the Pentagon’s chief technology officer, put it plainly: “At some level, you have to trust your military to do the right thing.”

There is also a credibility question worth raising about Anthropic itself. Critics, including some within the Trump administration, have argued that Anthropic’s vocal stance on AI risk and regulation is not purely principled. The accusation of “regulatory capture” suggests that Anthropic’s push for oversight benefits them competitively: onerous AI regulation is easier for a well-funded incumbent to absorb than for smaller startups trying to catch up. In other words, Anthropic has a financial interest in being seen as the ethical AI company. That doesn’t make their argument wrong, but it’s context worth having.

Weigh both of those points against what follows.

So What Did OpenAI Actually Agree To?

This is where the story gets murkier.

The day Anthropic’s deal collapsed, OpenAI signed its own agreement with the Pentagon. Altman announced on X that they had secured protections against both mass domestic surveillance and autonomous weapons. Those were the exact same two red lines that had just gotten Anthropic declared a supply chain risk.

How is that possible? How does one company get punished for demanding something, while another gets rewarded for claiming to have the same thing?

The honest answer is that we don’t fully know, because neither contract has been made public.

What we do know is how Altman framed the agreement. He said the Pentagon “agrees with these principles, reflects them in law and policy, and we put them into our agreement.” Read that carefully. He is saying the Pentagon agreed not to violate existing law, which circles back to the same issue Anthropic raised about the limits of what current law actually prohibits.

During Anthropic's own negotiations, the company said they were offered similar-sounding language packaged with what it called "legalese that would allow those safeguards to be disregarded at will." The protections looked real on paper and meant nothing in practice.

OpenAI may have genuinely won meaningful concessions. Or they may have accepted that same ineffective legalese and chosen to frame it as a victory. OpenAI’s own employees appear to be asking that question. This week, messages written in chalk appeared outside the company’s San Francisco offices: “Where are your redlines?” “You must speak up.” “What are the safeguards?”

Neither contract is public. That is, for now, where the story sits.

Then Came Iran

Here is the detail that reframes everything.

Even as the Pentagon was publicly blacklisting Anthropic and declaring it a national security risk, the U.S. military was using Claude in its strikes on Iran. Multiple sources confirmed to CBS News and the Wall Street Journal that Claude was the primary AI tool used by U.S. Central Command, helping to assess intelligence, run simulated war games, and identify military targets in a campaign that struck over 1,000 targets in its first 24 hours.

The Pentagon banned the company. And kept using the product.

Why does this matter to a piece about domestic surveillance? Because it raises a question about how much weight to give official assurances. The government announced Anthropic was a supply chain risk, a designation designed to signal that no contractor should touch their technology, while simultaneously relying on that technology in one of the most significant military campaigns in recent memory.

How you interpret that contradiction is up to you. But it is a data point worth keeping in mind as you evaluate the assurances being made on the surveillance side.

The Bigger Picture

Step back from the corporate drama for a moment and consider what actually happened here.

The U.S. government moved to effectively shut down a major American company. Not for fraud, not for antitrust violations, not for any crime, but over a contractual disagreement about how its technology could be used. Dean Ball, a former Trump AI adviser, used the phrase “attempted corporate murder.” Others have compared the approach to how authoritarian governments pressure their domestic tech companies: align with the state’s interests, or face consequences.

The Pentagon and its supporters would frame it differently. Their position is that a private company attempted to place restrictions on the sovereign military decisions of the United States government, and was held accountable for that overreach.

Both characterizations are in circulation. The facts, at minimum, are these: Anthropic held a position, the government responded with one of the most severe actions ever taken against a domestic American tech company, and a competitor stepped in and got the contract. What that sequence means for the broader relationship between AI companies and government power is a question the industry and the public are now actively grappling with.

Amodei has been preparing his employees for exactly this kind of moment for years. He required new hires at Anthropic to read The Making of the Atomic Bomb, a history of the Manhattan Project, because he believed that AI would eventually become as consequential to national power as nuclear weapons, and that the people building it needed to understand the moral weight of that. The scientists who built the bomb eventually lost control of how it was used.

Whether the same trajectory plays out with AI remains to be seen.

What is clear is that the surveillance question is not going away. That battle is quieter and slower than a military strike, and it plays out not in war zones but in the data flowing off your phone every single day.

What You Can Do With This Information

This piece isn’t meant to send you spiraling. But it is meant to make sure you understand what is being decided right now, in your name, with your data, by people who are not asking for your input.

A few things worth knowing:

The Fourth Amendment Is Not For Sale Act passed the House in 2024 with bipartisan support. It would require the government to obtain a warrant before purchasing Americans’ data from data brokers. It has not yet passed the Senate. If this issue matters to you, that is the specific legislation worth following.

You can also audit the apps on your phone. Location permissions are the most significant exposure. Any app that doesn’t functionally require your location to work probably doesn’t need it. Revoking those permissions doesn’t solve the larger problem, but it narrows your individual footprint in this system.

Most importantly: pay attention to this space. The decisions being made right now about how AI intersects with surveillance, military power, and civil liberties are being made fast, without public debate, and largely out of view. The Anthropic/Pentagon standoff broke through the noise briefly. Most of what comes next won’t.

You deserve to know what’s being built, and who it’s being built for.

A Note From Me

I try to give you the facts and the context and let you draw your own conclusions. That’s what this publication is for.

But you asked, or you will ask, where I stand. So here it is.

I’m not troubled by the military using AI. I’m not even troubled by the idea that the Pentagon needs operational flexibility. What I am troubled by is the specific combination of factors this story surfaces: powerful AI tools, a government with demonstrated interest in monitoring people’s digital lives, no national privacy law, and a legal framework where purchasing your data requires no warrant and leaves no trace.

The domestic surveillance question is not hypothetical. The infrastructure is built. The data is already being bought. What’s being negotiated right now is whether the most powerful analytical tools ever created get added to that system with no meaningful restrictions, because the law, as written, wouldn’t prohibit it.

That concerns me. Regardless of which administration is in power, regardless of who the target is today, surveillance infrastructure doesn’t come with a guarantee about who it gets pointed at tomorrow.

That’s why I cover this space. And that’s why I think it matters that you understand it.

— Allison

Safe Online Futures covers the technology stories that affect your everyday life. Subscribe to stay informed.