Artificial intelligence is moving from predicting trends and recommendations to a far more controversial frontier: predicting potential threats. The FBI is now exploring an AI-powered system that could analyze massive government databases and identify people who may require scrutiny before any crime takes place. But as the agency looks to automate its threat-screening efforts, critics are raising a major question — can an algorithm accurately predict danger, or could it create a new era of surveillance built on mistakes and assumptions?
The FBI’s shift to AI threat scoring
In a document published this past March, the FBI’s Threat Screening Center — the team behind the nation’s terrorist watch list — unveiled a bold new plan straight out of a science fiction movie. Federal records obtained by the magazine Reason show the bureau is looking for tech companies to build an artificial intelligence system that can flag potential threats before anyone breaks the law.
The official request calls for “Predictive Modeling Using Enhanced Data.” In plain English: instead of human analysts spending hours connecting dots by hand, the FBI wants fast software to scan massive troves of data, spot patterns in human behavior, and automatically score people as potential risks.
While the watch list has always aimed to stop threats before they happen, the main difference now is scale. The current list has ballooned to nearly two million names, and the FBI wants AI to sort through them instantaneously.
What counts as a threat?
This push for predictive tech comes as the federal government expands who it monitors. At the start of President Donald Trump’s second term, fewer than 10,000 people on the watch list were American citizens. That number could soon climb under new federal rules.
The FBI’s request specifically points to National Security Presidential Memorandum-7, signed in September 2025. The directive tells federal agencies to target domestic extremism tied to “anti-Americanism, anti-capitalism, and anti-Christianity,” as well as views on race, gender, and migration.
Critics worry these broad categories leave too much room for interpretation, turning ordinary political opinions and free speech into red flags for an algorithm.
Where the FBI gets your data
You don’t need to be accused of a crime to end up in the system. To feed these predictive algorithms, the bureau needs huge amounts of personal information — and much of it is bought on the open market.
On March 18, FBI Director Kash Patel confirmed to the Senate Intelligence Committee that the bureau routinely purchases Americans’ private location histories and personal details from commercial data brokers. Patel also noted that the FBI’s collection of biometric data (like facial recognition and fingerprints) is growing fast.
Patel hasn’t hidden these tech-heavy ambitions. “I’ve got every major tech company in the world embedded in the FBI,” he told Fox News, explaining that partnering with Big Tech is the fastest way to get “instantaneous results.”
The risk of getting it wrong
Legal experts and privacy watchdogs warn that relying on AI to predict human behavior is a dangerous gamble, mostly because algorithms are only as good as the data fed into them.
Research from the Brennan Center for Justice shows that these predictive systems carry high risks of bias and unwarranted surveillance. Making matters worse, recent policy changes have removed mandatory bias-testing for AI software bought by the federal government.
Then there is the issue of basic accuracy:
- Flawed databases: Government Accountability Office audits have repeatedly found widespread errors in existing watch-list files.
- Legal troubles: The Supreme Court has already ruled against the FBI twice for wrongfully placing innocent people on watch lists.
Civil rights advocates warn that layering predictive AI over an already error-prone database won’t make us safer — it will just make it easier for a bad computer guess to follow an innocent person for life.
Source:
inkl