Congress punts FISA reform as AI supercharges surveillance risks

Lawmakers approved a 45-day extension of a warrantless surveillance authority, delaying broader reform while artificial intelligence tools expand the government's reach.
Congress has once again chosen delay over decision. Lawmakers passed a 45-day extension of one of the most expansive warrantless surveillance authorities in U.S. law, kicking the can on reforms that have been stalled for months. The extension buys time but does nothing to address the growing tension between national security powers and the rapidly evolving capabilities of artificial intelligence, which now threatens to turn already sweeping surveillance into something far more intrusive.
The vote, which came just hours before the existing authorization was set to expire, keeps alive the surveillance provision — widely believed to be Section 702 of the Foreign Intelligence Surveillance Act (FISA) — without any of the procedural guardrails that privacy advocates and some lawmakers have demanded. The short-term patch means the underlying legal framework remains untouched, even as the intelligence community and law enforcement continue to collect communications without a warrant, including those of Americans who happen to talk to foreign targets.
That tension has been present since Section 702 was first enacted in 2008. But the technological context has shifted dramatically. When the provision was last reauthorized in 2012 and again in 2017, data analysis was largely manual and slow. Today, artificial intelligence systems can sift through enormous volumes of intercepted communications, identifying patterns, extracting meaning, and flagging subjects with speed and scale that were unimaginable a decade ago.
The combination of warrantless collection and AI-powered analysis creates a new class of risk. The government can vacuum up data without a warrant and then use machine learning models to mine that data for insights about individuals who were never the original target. Privacy experts have warned that this creates a de facto mass surveillance system, with AI serving as the engine that makes bulk collection useful and therefore irresistible.
Congressional efforts to rein in these powers have repeatedly stalled. Bipartisan proposals to require a warrant before querying Section 702 databases for information about Americans have failed to gain enough support, opposed by the intelligence community and its allies in both parties who argue such a requirement would hinder counterterrorism and counterintelligence operations. The 45-day extension punts that debate into a new session, but the underlying dynamics remain unchanged.
Artificial intelligence doesn't just amplify the reach of existing surveillance — it also makes the legal arguments for warrants harder to sustain. When an AI system can trawl through years of intercepted data and surface a connection between a U.S. person and a foreign target, the government can claim the link itself is evidence. That circular logic threatens to hollow out the Fourth Amendment's requirement for probable cause, replacing it with algorithmic correlation.
The 45-day extension provides no opportunity to address these concerns. It is a blunt procedural stopgap, not a substantive policy fix. Lawmakers will return in the new year to the same unresolved questions: whether the surveillance authority should be reauthorized at all, what limits should apply, and how to account for the fact that AI now makes every query a potentially sweeping search.
Some members of Congress have acknowledged the AI dimension. A handful of hearings have touched on the intersection of machine learning and surveillance, but no legislation has emerged that specifically addresses the problem. The intelligence committees, which have jurisdiction over FISA, have been slow to take up the issue, in part because the technology is evolving faster than the legislative process can keep up.
The 45-day extension also creates uncertainty for the technology companies that are compelled to assist with surveillance. Under Section 702, firms like Google, Meta, and Microsoft must comply with directives to turn over data. The short-term patch extends that obligation, but the lack of long-term stability complicates planning and compliance. Companies have grown increasingly vocal about the legal risks of participating in warrantless collection, especially when AI models trained on user data could be compelled to reveal information about their customers.
For ordinary Americans, the implications are opaque but real. The surveillance system operates largely in secret: the Foreign Intelligence Surveillance Court reviews only the government's applications, and the public rarely sees the details of how collection actually works. AI makes that opacity worse. When an algorithm flags your communications as suspicious, you have no way to know why, no way to challenge the assessment, and no way to learn whether the original collection was lawful.
Civil liberties organizations have called on Congress to include AI-specific safeguards in any reauthorization bill. Proposals have included requiring a warrant for queries that use AI to identify U.S. persons, mandating transparency reports on the use of machine learning in surveillance, and creating an independent review board to audit algorithmic targeting. None of those ideas made it into the 45-day extension.
The clock is now ticking. When the extension expires — late February, if the count starts from the passage date — lawmakers will face the same choice they just avoided. The difference is that each month of delay allows the intelligence community to deploy more sophisticated AI tools, normalize warrantless analysis, and build precedents that will be hard to reverse.
Congress has punted on FISA reform before, but the stakes are higher this time. The combination of a sweeping legal authority and a rapidly advancing technology creates a surveillance capability that the authors of the original law could not have imagined. The 45-day extension is not a pause — it is a continuation of the status quo, with all the risks that status quo now carries.
The question is whether lawmakers will use the extra time to craft meaningful AI-aware reforms, or whether they will simply kick the can again. History suggests the latter is more likely. But the technology is moving too fast for inertia to be a safe option.
Staff Writer
Maya writes about AI research, natural language processing, and the business of machine learning.
Comments
Loading comments…



