FISA 702 and the AI Governance Gap

Congress has voted again to extend FISA Section 702. The familiar debate has been cycling since 2008: warrants against national security, privacy against intelligence access, oversight against operational speed.

Here is what is not yet on the table.

The surveillance authorities under debate were designed for human analysts making human decisions at human speed. That world is gone. Autonomous AI agents now collect, correlate, and act on data at a scale and velocity that makes the 702 framework look like it was written for carrier pigeons.

When an AI agent ingests telemetry from twelve sources, infers a pattern of life, and triggers an automated response, who authorized that collection? What was the legal basis? Where is the audit trail? Under current law the answer to all three is often: no one knows.

Months get spent arguing about whether a human analyst needs a warrant to query a database. Meanwhile, AI systems are running millions of those queries autonomously, with no governance receipt, no attestation of authority, and no mechanism to challenge the output.

FISA reform matters. The harder problem is that the entire oversight model assumes a human in the loop. Agentic AI removed that assumption, and policy has not caught up.

The governance infrastructure for autonomous systems is not a future problem. It is a current gap.

What remains on the table:

• The 1978 / 2008 framework assumes a human decision-maker. Autonomous agents are not in that framework.
• The reauthorization debate is procedural. The structural gap is not procedural.
• Oversight requires an artifact that the system cannot rewrite. The 702 process does not produce one.

The loop closed around an oversight function that was never instrumented.