AI is raising the stakes in the government surveillance debate
A controversial provision used by the federal government as a backdoor for spying on Americans is up for renewal in April. This time, however, analysts say that the renewal of the Foreign Intelligence Surveillance Act Section 702 is of even greater stakes thanks to artificial intelligence, which has the potential to supercharge the government’s mass surveillance capabilities as well as the national security priorities laid out by President Donald Trump last year.
In 1978, when Congress originally passed FISA, the act was meant to establish a system of oversight for the U.S.’s foreign surveillance activities, with the bill establishing the FISA court to provide some judicial oversight. In 2008, however, the bill was changed with the passage of Section 702 of FISA, which ostensibly empowered intelligence agencies with the ability to conduct mass surveillance of targeted noncitizens.
The problem with FISA, however, is that it de facto created a system in which the government can collect and store information about essentially any foreigner it believes may have “foreign intelligence information,” a term so broadly defined that it might apply to almost anyone, and which has been used as a back door for spying on American citizens as well.
Section 702 has empowered law enforcement and intelligence agencies with the legal authority to collect communications going to and from the U.S. or between any U.S. citizens and foreign nationals en masse. These communications can then be searched by agents in a process that does not require a warrant, because the searches are supposedly limited to foreign targets. In real terms, this means that agencies like the FBI are conducting hundreds of thousands of searches a year, and many of them are targeting American citizens.
Some of the most egregious targets of this power have included members of Congress, donors to political campaigns, and political activists, ranging from Jan. 6 rioters to Black Lives Matter protesters. Even people just using a VPN to protect their internet browsing history may have their communications scooped up by law enforcement under this power.
Section 702, which must be periodically renewed or else the provision will sunset, was last renewed in the spring of 2024, as part of the Reforming Intelligence and Securing America Act signed by President Joe Biden. It introduced some reforms, the main one being a requirement that agencies acquire a warrant when searching 702 material for the exclusive purpose of collecting evidence of a crime. This accounts for a small fraction of all searches.
“This is an incredibly dangerous authority to have as unrestricted as it is.”
The current version of FISA, however, is set to sunset on April 20, unless Congress acts to renew it. Analysts are saying the renewal has even higher stakes this time, given the potential that AI has to supercharge abuses under FISA with less accountability and inflame the politically-charged national security priorities of the Trump administration.
“This is an incredibly dangerous authority to have as unrestricted as it is,” said Matthew Guariglia, a senior policy analyst at the Electronic Frontier Foundation. “The entire current national security apparatus seems poised to leverage all of the mechanisms of the state to go after people because of their First Amendment-protected activities. And now is not the time to be expanding those powers.”
The key difference with AI, according to Guariglia, is that where once a law enforcement officer had to sift through the data compiled under 702, now an AI program can do the grunt work, allowing agents to run queries and assemble information exponentially faster at the risk of introducing errors known as “hallucinations.”
“A generation ago, all this collected data, somebody would have had to go through it. Somebody would have had to read it and analyze it, and make inferences from it,” Guariglia said. “Now it’s being read by a computer, and those computers are making decisions or supposedly finding patterns, and they claim this is going to be a jumping-off point for intelligence and law enforcement. But what we’re seeing more and more is this becomes the judge, jury and executioner.”
We need your help to stay independent
The integration of AI queries into law enforcement is only a fraction of how the tech is being used in law enforcement, with Guariglia citing image search, facial recognition and automatic license plate readers as other examples of where new technology, built to handle large datasets, is impacting intelligence and law enforcement.
All of the new technology available to law enforcement is also colliding with Trump’s national security agenda, laid out in his seventh National Security Presidential Memorandum. The agenda, according to the document, includes targeting anti-fascist movements like antifa, as well as anything that the government considers supporting anti-Americanism, anti-capitalism and anti-Christianity. It also declares that people supporting any of these ideologies should be regarded as domestic terrorists.
In an interview with Salon, Patrick Eddington, a former CIA analyst and senior fellow at the Cato Institute, described a method by which a law enforcement agent could even use the new AI technology to essentially drum up a reason to open an investigation into anyone, leaning on both AI queries and the priorities laid out in NSPM-7.
Under the framework put forth by Eddington, a single FBI agent could open an “assessment,” the lowest level of an investigation, under the priorities described in NSPM-7, as well as the Attorney General’s guidelines for domestic FBI operations. These assessments require only an “authorized purpose” to open, a bar that can be met with no particular factual information, especially under the broad framework of NSPM-7. In Eddington’s example, a variety of people could be swept up in this dragnet if the hypothetical target attended an immigration enforcement protest, meaning the justification would be akin to those used to target protesters in the past.
After opening an assessment, an FBI agent could use AI queries of open source and databroker-provided information to compile a dossier on the target to meet the bar to open a preliminary investigation, which requires only that an agent show that a federal crime “may be” occurring.
In order to open a full field investigation into a target, an agent would then need to establish an “articulable factual basis” for belief that a federal crime may be occurring. While this is a higher bar to meet than any of the previous ones, former Attorney General Pam Bondi has created an infrastructure to make it easier to meet with her tip line system. Bondi was fired for unrelated reasons on April 2.
In a December Department of Justice memo, Bondi ordered the FBI to revise its tip line standards to allow “citizen journalists” to submit tips on “suspected acts of domestic terrorism.” The memo used the “doxxing” of law enforcement personnel like Immigration and Customs Enforcement officers as an example of the sort of behavior they were attempting to target, though again, under NSPM-7, belief in “radical gender ideology, anti-Americanism, anti-capitalism, or anti-Christianity” could also be seen as domestic terrorism.
If an agent is able to meet this factual basis, which could be satisfied by only the existence of a tip, even if it is unreliable, the agent would then unlock the ability to engage in everything from wire taps to FISA 702 queries against their target, who would at this point still have no knowledge of the investigation.
The guardrails that exist to prevent this process, dubbed “predicate laundering,” have, in the past, been shown to be reasonably effective. However, the Trump administration has taken efforts to prevent them from functioning. For example, the Privacy and Civil Liberties Oversight Board, a bipartisan board that is supposed to guard privacy, has been without a quorum since January 2025 and Inspector General Don Berthiaume at the DOJ is a Trump appointee.
“There’s a bureaucratic incentive in the FBI already to do this kind of stuff. The agents get rated on the number of assessments they open, the number of preliminary investigations, fulfilled investigations, enterprise-level investigations and the number of confidential informants they recruit,” Eddington said. “In an environment where you’re fighting for a promotion … It’s just a further incentive to agents to engage in this kind of activity. So it’s pernicious on multiple levels.”
Start your day with essential news from Salon.
Sign up for our free morning newsletter, Crash Course.
Eddington did, however, say that he believes there is hope that Congress will rein in some of the investigative powers unlocked under Section 702 because, in his assessment, there is not enough broad support for renewing the section in Congress to succeed at the moment, even despite Trump supporting the renewal. Trump had previously sought to kill the provision in 2024, citing at the time the well-known investigation into a Trump campaign advisor.
Part of the reason, however, is that a contingent of Republicans wants to pass the SAVE America Act, a voting restriction bill that could impact the midterm elections, before any other bill is passed. For now, that means the fate of Section 702 could be tied to that of the bill.
“There’s going to be a lot of wasted energy that goes into this. But as long as that particular group of Republicans continues to insist on moving the SAVE Act before they even think about doing anything else, for those of us who are trying to ensure that there is a real debate and that real alternatives to this actually get considered on the floor, that’s good news,” Eddington said.
Neither the FBI, NSA nor DOJ immediately responded to Salon’s request for comment.
Read more
about mass surveillance
The post AI is raising the stakes in the government surveillance debate appeared first on Salon.com.