The Fourth Amendment Gap in the National AI Framework

Recent artificial intelligence guidance fails to address the government’s use of AI for searches.

In March 2026, the Trump Administration published its National Policy Framework for Artificial Intelligence. The four-page document of legislative recommendations covers children’s safety, intellectual property, innovation, workforce development, electricity rates for data centers, and the federal preemption of state AI laws. It never mentions the Fourth Amendment.

A month later, the U.S. Supreme Court did. In Chatrie v. United States, the Justices heard the first case testing whether geofence warrants violate the Fourth Amendment. A geofence warrant is a legal order for a technology company to comb every users’ location data to identify the devices that were near a crime scene. The U.S. Court of Appeals for the Fifth Circuit has held that these warrants are unconstitutional because they are “general warrants”—the same open-ended writs that colonial British officers once employed against the American colonists. The constitutional framers wrote the Fourth Amendment to forbid general warrants because they do not identify specific subjects to be searched and could be used to search anyone. The U.S. Court of Appeals for the Fourth Circuit, reviewing the same question en banc, could not agree. The court split evenly, seven to seven, on whether collecting the data was even a “search,” and affirmed the appealed decision without a majority rationale.

Geofence warrants are just one kind of “reverse search.” These searches begin with a place, an image, a search term, or a behavioral pattern, then work backward to identify people from population-wide datasets. The split between the Fourth and Fifth Circuits shows how hard they are to regulate. While a court may focus on the single technology at issue in a particular case, executive action could set standards for the entire category of reverse searches. That is the gap that the recent AI framework could fill but does not; it sets no rules for reverse searches as a category.

The framework is explicit about what AI governance is for. It protects children from AI-generated abusive material. It urges the U.S. Congress to stop the government from coercing AI platforms to restrict speech based on political ideology. It opposes any new federal AI rulemaking body. What it does not address is the constraint aimed at the government itself: the limits the Fourth Amendment places on how the government uses AI to search. The same document that invokes the First Amendment to check government overreach through private platforms says nothing about government overreach through search warrants.

Chatrie itself may not settle the issue. The case concerns a 2019 search that relied on Google location data. In 2023, Google migrated its users’ location histories off of its centralized “Sensorvault” database and onto users’ individual devices. A geofence warrant works by compelling a company to search its own database—move that data off of the database, and there is little left for a geofence warrant directed at Google to reach. That change does nothing, however, about keyword warrants, facial recognition—used by at least 18 of 24 federal agencies surveyed in 2020—predictive policing algorithms, or other reverse-search tools.

Commentators expect a narrow ruling in Chatrie, perhaps one resting on the defendant’s decision to opt in to Google Location History. That question covers one company and one crime. The question it sidesteps is larger: May the government use reverse-search warrants across AI systems with no direct statutory rules restricting those warrants?

The legal landscape that the framework inherits offers little protection. Congress passed the Electronic Communications Privacy Act in 1986, before smartphones or location databases existed. President Joseph R. Biden’s two relevant executive orders—Executive Order 14074 on policing reform and Executive Order 14110 on AI governance—were both rescinded at the start of the second Trump Administration. The new framework advocates for a federal AI law that preempts state AI laws, yet it sets no federal floor for law enforcement’s use of AI. States that have restricted geofence warrants or paused facial recognition now face the prospect of preemption without nothing put in its place. Neither had the rescinded Biden-era orders set a statutory standard. A durable federal floor has to come from Congress.

What that floor should contain is not a mystery. Model legislation already supplies workable language: escalating judicial approval required at each stage of a reverse search, limits confining these warrants to serious felonies, and mandatory deletion of data drawn from unlawful collection. The Fourth Amendment Is Not For Sale Act passed the U.S. House of Representatives 219 to 199 with bipartisan support, which shows that statutory particularity standards for government data access are politically viable. Congress does not need the Supreme Court’s permission to recognize that AI-powered dragnet searches raise constitutional problems.

The White House’s framework recognizes that AI is too consequential to leave to 50 state legislatures working without coordination. Then, incongruent with that principle, it says nothing about the constitutional limits on government surveillance. A policy document that uses the First Amendment to limit what the government may ask technology companies to do, but no the Fourth Amendment to limit what data the government may order them to produce, is an incomplete framework for AI governance.

Ali Feroz Ansari is a J.D. candidate at the Sandra Day O’Connor College of Law at Arizona State University