CIA Uses GPT-4 to Generate Intelligence Reports for Police

The Central Intelligence Agency has been producing intelligence analyses using GPT-4 and distributing them to local law enforcement agencies through the Department of Homeland Security's fusion center network, according to documents obtained through a public records request.

The CIA's Open Source Enterprise (OSE) has been creating daily reports by feeding international news articles into commercially available generative AI tools—specifically GPT-4.1, the same technology that powers ChatGPT. The agency uses keyword searches and relevancy algorithms to select articles from major outlets in foreign countries, then employs the AI to refine selections, translate headlines, and summarize content. CIA officers review the outputs at each stage to validate accuracy.

Documents obtained by Reason through Washington state's Public Records Act show the OSE generated daily security situation reports covering Canada and Mexico during this year's soccer World Cup. These reports were then passed to the Washington State Fusion Center, part of a nationwide intelligence-sharing network established after the September 11 attacks.

Why it matters

This represents a significant expansion of AI into the intelligence-to-policing pipeline. While the reports themselves contain no classified information and focus on publicly available sources, they demonstrate how federal agencies are operationalizing large language models for routine intelligence work—and how that AI-generated analysis flows directly to local law enforcement. The fusion center network has faced persistent criticism for inefficiency and distributing flawed information, raising questions about quality control when AI enters the equation.

From chatbot to operational tool

CIA officials have been signaling this shift for several years. OSE Director Randy Nixon told Bloomberg in 2023 that the agency was developing a chatbot for intelligence analysts. By April 2026, CIA Deputy Director Michael Ellis announced the agency had produced its first entirely AI-generated report.

The methodology notes accompanying the World Cup reports are explicit about the process: OSE analysts are essentially using ChatGPT as an intermediary between foreign news coverage and domestic law enforcement briefings. The reports themselves don't contain shocking revelations—they represent "open source intelligence," the longstanding practice of mining public media for security indicators.

Broader AI integration in national security

The CIA's use of GPT-4 for routine reporting fits within a broader pattern of AI adoption across U.S. security agencies. Immigration and Customs Enforcement employs AI for facial recognition and real-time individual tracking, drawing on massive datasets from government and private sources. The FBI has solicited contractors to build AI software for maintaining watch lists.

Most significantly, the military uses AI targeting systems like Project Maven to identify enemy targets at speeds beyond human capability. While early speculation suggested AI may have been responsible for the Minab school bombing that killed dozens of Iranian schoolchildren, subsequent investigation pointed to human error instead.

What remains unclear from the documents is how local police departments actually use these AI-generated CIA reports. The Washington State Fusion Center redacted portions of the materials for security reasons, leaving the operational impact uncertain.

These details were first reported by Reason, which obtained the documents during an unrelated public records request for World Cup-related police bulletins.