Democrats can’t decide whether to hand Trump warrantless surveillance powers — and that tells you everything

When Jim Himes, the top Democrat on the House Intelligence Committee, announced his support for a clean reauthorization of Section 702 surveillance powers, Jamie Raskin — his counterpart on the Judiciary Committee — was already circulating a memo demanding fundamental reforms. The Democratic Whip’s office responded by sending both positions to the caucus without commentary, as though presenting a wine list rather than a party line on whether to hand Donald Trump one of the most powerful warrantless surveillance tools ever constructed. That institutional shrug tells you everything about why Congress is about to sleepwalk into authorizing a surveillance apparatus it no longer understands.

The core problem is not that Democrats disagree. Disagreement is normal. The problem is that this debate is being conducted using the vocabulary and assumptions of 2018, while the actual surveillance infrastructure has leapt into a fundamentally different technological era. Section 702, the legal authority that permits intelligence agencies to conduct warrantless surveillance of Americans’ communications, is up for reauthorization — and the framework Congress built to constrain it has been rendered obsolete by artificial intelligence, at precisely the moment the executive branch has gutted the oversight bodies meant to prevent abuse.

Jake Laperruque, deputy director of the Center for Democracy and Technology’s security and surveillance project, noted that Democrats appear uncertain about their position on the surveillance authority, and expressed hope that recent politicization at DOJ would influence their stance against reauthorization. That uncertainty is not a quirk of legislative sausage-making. It is the visible symptom of a Congress that has not grappled with what Section 702 actually means in 2026 — a year in which the surveillance machine runs on capabilities that did not exist when the current legal framework was written.

Here is what has changed. When a human analyst conducts a backdoor search on an American under Section 702, the search is logged, potentially reviewed, and limited by the analyst’s time and attention. Past reviews of the program have regularly found violations even under those constraints, including searches on Black Lives Matter activists and sitting members of the U.S. Senate. An inspector general review found that while apparent violations decreased following reforms enacted during the last reauthorization cycle, a secret court opinion reported by the New York Times found significant problems with how the government tracks its searches of information about Americans.

Now multiply that system by artificial intelligence. Kasten told The Intercept that intelligence agencies can now use AI to analyze vast amounts of data, a capability that could be beneficial or harmful depending on how it’s applied. That careful phrasing conceals an enormous shift. When an AI system conducts the kind of query that once required a human analyst, it can do so at scale — across millions of records, drawing connections that no person would have the bandwidth to find, flagging patterns across communications that no individual analyst would ever read. The oversight framework built for human-speed surveillance has no answer for machine-speed surveillance. The reforms Congress passed in 2024, however earnest, were designed for a world that no longer exists.

This technological leap would be alarming under any administration. Under this one, it is something closer to a structural emergency. The Trump administration has systematically dismantled the oversight bodies that were supposed to prevent abuse of Section 702 authorities. Independent inspectors general have been fired or sidelined. The Department of Justice has been explicitly politicized. The Privacy and Civil Liberties Oversight Board — the independent body Congress specifically created to review surveillance programs — has been hollowed out. The guardrails were already insufficient for the pre-AI era. Now the guardrails are being removed while the machine accelerates.

Centrist Democrats argue that the 2024 reforms were sufficient. Progressives counter that reforms are only as durable as the institutions enforcing them, and those institutions are under siege. But this framing — centrist versus progressive, hawk versus dove — misses the point. The question is not whether the old reforms were adequate for the old system. The question is whether any member of Congress can credibly claim to understand what they are authorizing when the surveillance capabilities have changed faster than the oversight architecture, and when the administration has demonstrated its willingness to weaponize executive power against political opponents.

Ron Wyden, the Oregon Democrat who has been the Senate’s most persistent critic of surveillance authorities, has argued that classified information about Section 702 should be made public and debated before any reauthorization occurs. Wyden’s framing matters because it points to a dimension of this fight that rarely gets airtime: classification as a tool of political control. When the government classifies the ways it uses surveillance authorities — including, crucially, the ways AI is now integrated into the collection and analysis pipeline — it prevents the public and most members of Congress from making informed judgments. The debate defaults to trust. Trust the intelligence community. Trust the courts. Trust the process. In an administration that has explicitly politicized the Justice Department and gutted independent oversight, the foundation of that trust has been removed.

Digital rights advocates report something new in this FISA fight: genuine anger from constituents who are not traditional civil liberties activists. Parents worried about their teenagers’ social media data being accessible to intelligence agencies. Small business owners concerned about commercial data brokers selling transaction data that ends up in government hands. The concerns are less abstract than in years past. People understand data now in a way they didn’t in 2018 or even 2024. And they understand, intuitively, that AI changes the equation — that a database is one thing when a person has to search it, and something entirely different when a machine can search all of it, all the time, looking for whatever it has been trained to find.

The progressive demand list reflects this shift. Reformers want warrant requirements for searches involving Americans. They want restrictions on the government’s ability to purchase commercially available data from brokers — a practice that effectively allows intelligence agencies to bypass warrant requirements by buying what they cannot legally collect. And they want AI-specific safeguards: limits on automated querying, transparency requirements for algorithmic analysis of Section 702 data, and mandatory audits of how machine learning models interact with the surveillance database. The precise form of those safeguards remains a matter of debate, but the underlying principle is simple — you cannot govern twenty-first-century surveillance with twentieth-century rules.

The Congressional Black Caucus sits in a particularly uncomfortable position. As the American Prospect reported, the caucus faces pressure to support reauthorization even though Section 702 authorities have been used to surveil Black Lives Matter activists. The historical pattern is long and well-documented. From COINTELPRO to the post-Ferguson FBI investigations of protest leaders, surveillance powers have been disproportionately aimed at Black political organizing. Now add AI-driven pattern recognition to a system already prone to racial bias in targeting, and the risk compounds in ways that no procedural reform can adequately address. To vote for a clean reauthorization is to affirm a system that has been used against the communities these members represent — and to trust that AI will not deepen those abuses. To vote against it is to accept the political risk that comes with being labeled as weak on national security.

Political strategists describe the CBC’s dilemma as a case study in how the committee system itself shapes ideology on surveillance questions. A member who has served on the Intelligence Committee for years develops relationships with intelligence officials, absorbs their framing of threats, and gradually internalizes the view that these authorities are necessary. This dynamic can be more powerful than any lobbying campaign. But it also means that the members closest to the intelligence community are the ones least likely to grasp — or to challenge — how fundamentally AI has altered what Section 702 enables. They are being socialized into defending a program whose operational reality has shifted beneath their feet.

This connects to a broader pattern in how institutional relationships shape decision-making across industries. But in the surveillance context, the stakes are categorically different. The asymmetry of political risk — where the danger of restricting surveillance is concentrated and personal, while the danger of enabling abuse is diffuse and institutional — now operates on top of a technological asymmetry that Congress has barely begun to acknowledge. AI does not just make surveillance more efficient. It makes surveillance qualitatively different: more pervasive, more connective, more opaque, and far harder to audit after the fact.

The silence from Democratic leadership is the sound of a party that knows the old arguments for reauthorization no longer hold, but has not built the political infrastructure to make new ones. Himes and the centrists are arguing from precedent — every previous FISA fight ended in reauthorization, and the sky did not fall. Raskin and the reformers are arguing from principle — that an authoritarian executive cannot be trusted with unchecked power. But neither camp is adequately reckoning with the argument from technology: that what Congress is being asked to reauthorize in 2026 is not the same program it authorized in 2018, or renewed in 2024. The legal text may be identical. The operational reality — AI-augmented, at machine scale, with diminished oversight — is not.

If Democrats reauthorize Section 702 without AI-specific safeguards, they will have voted to extend a legal framework designed for targeted foreign intelligence collection into an era of automated mass analysis — under an administration that has already demonstrated its willingness to use executive power against political opponents, journalists, and activists. If they block reauthorization without offering an alternative, they will be blamed for every intelligence failure, real or invented, that follows. The difficult middle path — genuine reform that addresses the AI transformation while preserving legitimate intelligence capabilities — requires a level of technological literacy and political courage that this Congress has not yet demonstrated. The vote will tell us whether it can.