FBI Seeks Predictive AI for Federal Terrorism Watchlist

FBI Explores AI-Powered Threat Prediction

The FBI is seeking artificial intelligence technology capable of searching across federal databases and generating predictive models for its Threat Screening Center, the office responsible for maintaining the government's consolidated terrorism watchlist. A March 27 procurement request posted to SAM.gov outlines specifications for a system that would fundamentally change how the bureau identifies potential threats.

The proposed technology would create an AI-enabled knowledge base that searches databases held across different government systems, summarizes records, connects identifiers, and answers questions in natural language while retaining citations. Most significantly, the system would perform "predictive modeling using enhanced data with traceable lineage"—comparing new information against existing records to identify correlations and predict where investigators might find additional relevant material.

While the request does not indicate the FBI has purchased or deployed such a system, the specifications reveal the capabilities the bureau aims to develop. The technology would not independently place individuals on the watchlist, but would instead recommend connections for human analysts to review.

Why it matters

This moves predictive policing from local crime forecasting into a federal system that affects travel, surveillance, and law enforcement encounters nationwide. Federal oversight agencies have already documented accuracy problems, misidentifications, and limited redress options in the current watchlist system. Adding AI-generated correlations introduces another layer of automated decision-making into a process that can restrict civil liberties based on classified information individuals cannot contest. The concern intensifies given recent policy directives expanding watchlist criteria to include broader categories of domestic threats tied to political beliefs.

Accuracy Problems in Existing Systems

Traditional predictive policing programs have produced troubling results. An examination of 23,631 forecasts made by Geolitica (formerly PredPol) for Plainfield, New Jersey, found fewer than 100 matched a reported crime of the predicted type in the designated area and timeframe—a success rate below 0.5 percent.

The FBI's proposed system differs from location-based crime forecasting, but faces similar structural challenges. Predictive systems inherit weaknesses from underlying records, which contain mistaken identities, incomplete reports, and data generated through previous enforcement decisions. If databases already reflect disproportionate scrutiny of particular communities, AI may interpret that concentration as evidence of higher threat levels, creating a self-reinforcing cycle.

Watchlist Scope and Oversight Gaps

The Threat Screening Center was created after September 11 to consolidate terrorism databases. It now shares information with federal agencies and state, local, tribal, and international law enforcement. Officers can receive watchlist alerts during traffic stops and contact the center for instructions without informing the individual.

The Privacy and Civil Liberties Oversight Board reported in 2025 that the list contained approximately 1.1 million people, including fewer than 6,000 U.S. persons. A Government Accountability Office review of 289 watchlist-related inquiries filed by U.S. persons between December 2021 and September 2023 found 21 cases of misidentification, 88 removals, and nine placements on less restrictive subsets. A January 2026 GAO report found the FBI had not ensured state and local users understood watchlist policies or received adequate training.

Unanswered Questions

The FBI has not disclosed what data the proposed technology would examine, what its models would predict, how the system would be tested for false matches, or whether individuals could learn that automated connections contributed to their scrutiny. These omissions are particularly concerning in a system federal reviews have found includes misidentifications, outdated records, and limited redress avenues.

The danger is not necessarily that algorithms will independently add names to the watchlist, but that AI-generated correlations will begin functioning as evidence before the public understands what rules govern their use. Details of the procurement request were first reported by Military.com.