The legislation would require U.S. officials to report how AI tools access and process sensitive surveillance data, increasing oversight of national security systems.

Senator Adam Schiff (D-Calif.) introduced S.4402 to require federal agencies to disclose how they use AI in surveillance, aiming to provide Congress with visibility into how sensitive intelligence data is analyzed.

The bill directs the Attorney General and the Director of National Intelligence to produce a report within 120 days. Agencies must submit a report to Congress and the Foreign Intelligence Surveillance Court (FISC). It must also include a public version with classified details removed.

The report must identify each AI system used with data collected under the Foreign Intelligence Surveillance Act (FISA). The report must explain what each system does in practice, including how AI is used to review, sort, or analyze surveillance data.

The bill requires reporting in three key areas:

Systems training. Federal agencies must disclose what data they use to train AI systems. They must state whether the systems can access unminimized data, which includes raw communications that may contain information about U.S. persons. Further, agencies must describe any limits placed on that access.

Testing and monitoring. The bill requires agencies to explain how performance is evaluated and how errors are identified. It must also state whether FISC reviewed or approved the use of these systems.

Notification of FISA data access. Agencies must inform Congress and FISC before allowing a new AI system to access FISA-collected data. This applies each time a new system is introduced into the process.

The measure is limited to disclosure obligations. It does not change how surveillance is conducted, but only focuses on how AI is used within existing intelligence operations.