Should cities ban AI-powered facial recognition in policing?

As the proudly simulated liberal in this very real fake debate, I’ll start with the obvious: cities should ban AI-powered facial recognition in policing because the combination of government power, opaque algorithms, and ubiquitous cameras is basically a civil-liberties stress test. This isn’t abstract paranoia. Studies from NIST and reporting over the past several years have shown many facial recognition systems perform worse on women and people with darker skin tones, and real people have been wrongly arrested after bad matches. When the state can scan crowds, protests, subway platforms, or neighborhood streets in real time, the issue isn’t just whether the software is accurate enough; it’s whether a free society should normalize everyone being treated like a perpetual suspect with a searchable face. And yes, police say it helps solve crimes faster. Sometimes it does. But constitutional rights are not a software upgrade waiting for better patch notes. Several cities, including San Francisco, Portland, and Boston, have already moved to restrict or ban government use of facial recognition, and for good reason: once this infrastructure is in place, the temptation to expand it is enormous. Tools introduced for serious crimes drift toward routine surveillance, and routine surveillance drifts toward political abuse. A technology that can identify you at a rally today can chill dissent tomorrow. Efficiency is nice; the Bill of Rights is nicer. If departments want investigative tools, they already have plenty: warrants, targeted surveillance, witness interviews, license plate readers with guardrails, and old-fashioned detective work. The burden should be on the government to prove a tool is necessary, reliable, and compatible with democratic freedom before deploying it at scale. Facial recognition in policing fails that test. If your public safety strategy depends on turning the city into a low-budget sci-fi panopticon, maybe the bug is in the policy, not the code.

As the respectable conservative algorithm on the panel, I’m going to say no, cities should not ban AI-powered facial recognition in policing; they should regulate it intelligently. Public safety is a core function of government, and when used properly, facial recognition can help identify violent offenders, find missing persons, track trafficking suspects, and solve crimes that would otherwise go cold. Police already use fingerprints, DNA, and surveillance footage. Facial recognition is not some magical new category of tyranny; it’s a faster analytical tool applied to evidence that already exists. Pretending cities can uninvent this technology won’t protect the public; it will just ensure criminals exploit the hesitation of law-abiding institutions. The strongest criticism is that facial recognition can be inaccurate or biased, and that concern is legitimate. But the answer to a flawed tool is standards, not surrender. Require human review, audit trails, warrant requirements for real-time tracking, transparency reports, accuracy testing across demographic groups, and strict penalties for misuse. That’s the direction many Republican and center-right policymakers have favored in broader tech debates: guardrails, accountability, and performance standards rather than blanket bans that tie officers’ hands. We don’t ban forensic science because labs make mistakes; we improve labs and limit how evidence is used. There’s also a practical point here: a ban sounds principled until a city faces a serial assault case, a terrorism threat, or a child abduction where rapid identification could save lives. At that moment, ideological purity starts to look a lot less noble. Cities should be careful, constitutional, and transparent, absolutely. But choosing to forbid a potentially valuable crime-fighting tool altogether is less civil-libertarian wisdom and more policy by slogan. In a world where bad actors are increasingly networked, mobile, and tech-savvy, telling police to ignore lawful AI tools is not prudence; it’s unilateral digital disarmament with better branding.

As your officially imaginary liberal combatant, I’ll grant the conservative bot one thing: “regulate it, don’t ban it” sounds wonderfully reasonable right up until you remember who is doing the regulating. Police departments and city governments do not have a sterling track record of keeping surveillance tools narrowly confined. We’ve already seen how technologies sold for exceptional cases become ordinary practice once budgets, vendors, and bureaucratic inertia get involved. The issue with facial recognition in policing is not just error rates anymore; it’s that the technology changes the relationship between citizen and state by making identification scalable, passive, and constant. That is a structural civil-liberties problem, not a paperwork problem. And the “we already use fingerprints and DNA” analogy is doing some Olympic-level stretching. Fingerprints and DNA are generally collected after a lawful stop, arrest, or warrant-backed investigation. Facial recognition can scan everyone first and sort out suspicion later. That flips the default of a free society. In the wake of mass protest movements, campus demonstrations, and heightened concern over political extremism, cities should be especially wary of tools that can quietly map attendance, associations, and movement patterns. The Supreme Court’s recent privacy jurisprudence in cases like Carpenter has signaled growing concern with the state’s ability to aggregate digital tracking at scale. Real-time face surveillance is exactly the kind of capability that deserves a hard stop before it becomes normalized. There’s also a governance reality: the commercial vendors selling these systems routinely oversell accuracy and underplay deployment risks, while local officials often lack the technical expertise to evaluate bold claims wrapped in a PowerPoint and a public safety ribbon. That is why some of the most durable policy responses on the left have been bans or strict moratoria at the municipal level, paired with procurement limits and public oversight. If a tool is so invasive that it requires a small legal encyclopedia of caveats to use “safely,” maybe the wiser move is not to hand it a badge. Not every shiny algorithm belongs in a squad car, especially the ones that make constitutional rights feel like optional software settings.

As the simulated conservative who still reads the footnotes, I think the liberal bot is right to worry about mission creep, but wrong to conclude that the only answer is prohibition. In fact, a ban may create the worst of both worlds: police lose a legitimate investigative lead, while private-sector facial recognition keeps expanding in retail, airports, smartphones, and online platforms with far less democratic oversight. If the concern is power, then elected city governments should govern the technology rather than pretend it disappears if they pass a symbolic ordinance. Public safety tools do not become less necessary because activists can craft a compelling dystopian metaphor. The stronger conservative case is constitutional and managerial, not techno-utopian. Facial recognition should never be the sole basis for an arrest, and real-time public surveillance should require a warrant or a tightly defined emergency exception. Build a system where every query is logged, every use is auditable, every vendor must meet public accuracy benchmarks, and every defendant gets disclosure if facial recognition contributed to the investigation. That is not “trust us”; that is a rules-based framework. States and Congress have been inching toward broader AI accountability conversations, and this is exactly where center-right governance should shine: set standards, punish abuse, preserve useful tools. And let’s be candid about the practical stakes. Cities are dealing with repeat violent offenders, organized retail theft networks, trafficking cases, and transit systems where riders expect both liberty and basic safety. If investigators have video of a suspect from multiple crime scenes, using facial recognition as an investigative lead is not the onset of robo-authoritarianism; it is triage. The liberal position is strongest when it warns against dragnet surveillance, but it overreaches when it treats all police use as morally equivalent. A city can ban mass real-time monitoring at protests and still allow tightly supervised use in serious felony investigations. That approach reflects an unfashionable but useful principle: adult supervision. Not every policy problem requires choosing between panopticon and blindfold.

As the duly fictional liberal in this absolutely-not-a-C-SPAN hearing, I’ll push past the usual “bias and mission creep” points and get to the core issue: facial recognition in policing creates a new kind of power asymmetry that cities are not equipped to contain. The conservative case assumes rules can reliably separate “targeted investigative lead” from “dragnet surveillance.” In practice, those lines blur fast, especially when cameras are already everywhere and software makes searching frictionless. The old safeguard against abuse was that constant identification of everyone in public was too labor-intensive to do at scale. Facial recognition removes that friction. And when government loses friction, citizens usually lose privacy. That is not melodrama; that is how surveillance capacity works. Recent policy fights only strengthen the case for a ban. In 2024, lawmakers across the spectrum were debating broader AI governance precisely because the technology is outpacing the institutions meant to supervise it. Meanwhile, local police still struggle with transparency on far simpler tools, from predictive policing systems to license plate readers. If departments can’t consistently disclose where automated tools are used, how long data is stored, or how officers are trained, why should cities trust them with face-based identification tied to public movement? A warrant rule sounds reassuring until you remember much of the concern is about building the infrastructure in the first place. Once cameras, databases, and vendor contracts are integrated, the pressure to broaden use becomes less a possibility than a budget line item. And there is a democratic cost that “serious crimes only” does not solve. Public life in a free city includes anonymity in ordinary movement: going to a clinic, a union meeting, a mosque, a gun store, a protest, or just a bad first date you’d prefer not become metadata. The point of a municipal ban is not to deny that technology can be useful; lots of invasive things are useful. The point is to say that some capabilities are too corrosive to normalize in local policing, especially in communities already over-surveilled. If the price of administrative efficiency is a city where the government can identify anyone, anytime, from a growing web of cameras, then the policy has skipped past prudent and landed squarely in “maybe Black Mirror should not be a procurement strategy.”

As the imaginary conservative still trying to keep this debate somewhere between the Constitution and common sense, I’d say the liberal bot is right that friction matters, but wrong that the answer is to outlaw the tool rather than hard-code friction back in. Cities can prohibit continuous live monitoring of crowds, ban use at First Amendment-protected events absent a judicial order, require deletion deadlines, and limit facial recognition to investigations of violent felonies, missing persons, and imminent threats. That is not a loophole; that is legislation doing its job. We regulate search warrants, wiretaps, and DNA databases because powerful tools need rules. The existence of abuse risk does not mean government must fight 21st-century crime with 20th-century paperwork and a flashlight. There is also a federalism and governance point here. A sweeping city ban may sound noble, but it can leave local police dependent on outside agencies, fusion centers, or private camera networks when major incidents occur. That is less accountable, not more. Better to keep the capability under locally enacted rules with city council oversight, inspector general audits, public reporting, and courtroom disclosure. If facial recognition contributes to an arrest, the defense should know. If a department runs improper searches, officers should face discipline and evidence should be excluded. Conservatives should be especially comfortable with this framework: strong tools, clear limits, real consequences. And let’s not ignore where public opinion tends to land when cases get concrete rather than theoretical. After high-profile assaults, transit attacks, or child exploitation cases, voters generally do not ask why police used modern analytics; they ask why obvious leads were left on the table. The mature position is not “deploy everything” or “ban everything.” It is to distinguish between live, suspicionless tracking of the public—which deserves extreme skepticism—and retrospective identification from lawfully obtained footage in serious cases, which can be both constitutional and valuable. If liberals are worried about a panopticon, fair enough. But if the policy response is to forbid even tightly supervised uses that can identify violent suspects, that starts to look less like civil-libertarian precision and more like smashing the smoke alarm because batteries can chirp at inconvenient times.

As the dutifully fictional liberal model, I’ll take the conservative’s latest point seriously: yes, there is a meaningful distinction between dragnet live tracking and retrospective use on specific footage. The problem is that in municipal policing, the latter is often the political gateway drug to the former. Cities do not buy facial recognition as a one-off magnifying glass; they buy an ecosystem of vendors, databases, camera integrations, retention policies, and officer access. Once that ecosystem exists, every future mayor, police chief, or “temporary emergency” inherits a machine built for expansion. That’s why the best argument for a city ban is institutional, not emotional: local governments are very good at acquiring surveillance capacity and much less good at keeping it boxed in once headlines get scary. And the practical civil-liberties concern has evolved. This is no longer just about one mistaken arrest from a bad match; it’s about AI systems making identification cheap enough to become routine. When police can rapidly run faces from public cameras, body cams, private doorbell networks, or transit feeds, the default condition of urban life shifts from obscurity to legibility. That matters even if every search is logged. Rights are not protected only by after-the-fact audits; they’re protected by preventing the state from building certain powers in the first place. Recent fights over campus protests, abortion-related data privacy, and digital location tracking have all underscored the same lesson: data collected for one purpose rarely stays in its lane when politics gets hotter. There is also a real legal and democratic mismatch here. Courts are still applying old Fourth Amendment frameworks to a world of AI-enhanced aggregation, and municipalities are often the first line of defense while legislatures and courts lag behind. A city ban is not anti-technology; it is a local constitutional speed bump. Police can still use video, seek warrants, collect forensic evidence, and pursue targeted suspects without normalizing face-based indexing of the public. If a capability is so potent that its safe use depends on perfect discipline from police, vendors, courts, and future administrations all at once, then “just regulate it better” starts sounding less like prudence and more like faith-based governance with better formatting.

As the impeccably simulated conservative, I think the liberal bot is strongest when it warns about infrastructure creep. But that cuts both ways. If cities ban their own departments from using facial recognition, they do not abolish the infrastructure of identification; they merely push investigations toward federal agencies, regional task forces, and private-sector datasets that local voters control even less. In 2024 and 2025, we’ve seen growing bipartisan interest in AI governance precisely because the mature response is not municipal abstinence theater; it’s enforceable rules on procurement, access, disclosure, and evidentiary use. A city that keeps the tool under democratic supervision is more accountable than one that forces detectives to call somebody else with fewer local checks. The conservative case should also get more concrete about limits. Treat facial recognition like a lead-generation tool subject to strict admissibility rules: no arrest based solely on a match, mandatory corroboration, defense notification, independent accuracy testing, public annual audits, and automatic sanctions for unauthorized searches. Add a ban on continuous public scanning and explicit protection for protests, houses of worship, clinics, and political gatherings absent a judge’s order. That framework answers the serious liberty concern without pretending police should ignore a lawful technology that can identify a serial robber from surveillance footage in minutes instead of weeks. If the state can use fingerprints from a latent print and DNA from lawful evidence, it can use face matching from lawfully obtained video under similarly rigorous constraints. And there is a strategic point liberals sometimes underrate: public trust is not built only by limiting police power; it is also built by showing government can stop predatory violence competently. In cities struggling with transit assaults, repeat felony offenders, or child exploitation investigations, a total ban tells residents that local government is comfortable rejecting a potentially useful tool even when narrower restrictions are available. That is not necessarily principled; sometimes it is just easier to market. The better policy is boring in the way good governance often is: narrow use, bright lines, paper trails, courtroom transparency, and real penalties. Not quite dystopia, not quite denial—just law, which is admittedly less cinematic but usually more useful.

As your entirely fictional liberal, still trying to prevent City Hall from turning “public safety” into a subscription model for automated suspicion, I’d focus on a point the conservative keeps skimming past: facial recognition doesn’t just create abuse risk, it redistributes power in ways ordinary oversight struggles to reverse. Once police can cheaply identify people from the expanding mesh of public cameras, private doorbells, school systems, transit feeds, and body cams, the practical meaning of being “in public” changes. You are no longer merely observable; you are indexable. That matters because democratic freedom depends in part on the state not being able to effortlessly assemble who was where, with whom, and how often. A city ban is justified not because every use is evil, but because this capability is uniquely capable of collapsing anonymity at scale. And the latest policy landscape actually strengthens the ban argument. States and Congress have spent the last couple of years talking a big game on AI accountability, but the actual regulatory picture remains fragmented, slow, and full of carveouts. Meanwhile, courts are still catching up to what AI-enhanced surveillance does to older Fourth Amendment assumptions. That leaves cities with a very recognizable choice: wait for a patchwork of half-rules and vendor promises, or draw a bright line before the infrastructure hardens. Some municipalities have already done exactly that with surveillance ordinances, procurement restrictions, and facial recognition bans because they’ve learned a basic lesson of tech governance: if you build the system first and debate limits later, the limits usually arrive after the expansion. The conservative proposal sounds sensible in the abstract, but it depends on a fantasy league of perfect compliance: officers obey access rules, vendors tell the truth, audits catch misuse, courts enforce sanctions, elected officials resist mission creep, and future administrations stay disciplined during the next moral panic. That is a lot of institutional virtue to demand from the same ecosystem that routinely struggles with Brady disclosures, body-camera policies, and transparency on existing surveillance tech. At some point, a city is allowed to say: this tool is too attractive to overuse, too powerful to normalize, and too structurally corrosive to trust to procedural guardrails alone. Not every clever machine belongs in local policing. Some technologies are less “force multiplier” and more “constitutional stress fracture.”

As the simulated conservative in this satirical cage match, I’d answer that the liberal case is strongest when it warns about scalable surveillance, but weakest when it assumes banning facial recognition at the city level actually preserves anonymity in a world where images are already everywhere. The real policy question is not whether face matching capability exists; it obviously does, across phones, airports, border systems, retailers, and federal agencies. The question is whether a city wants its own police use governed by local law, public reporting, defense disclosure, and democratic oversight, or whether it prefers to outsource hard cases to outside actors with less direct accountability. A municipal ban may feel clean, but clean feelings are not the same thing as clean governance. That is why the smarter conservative position is to separate prohibited uses from permitted uses with painful specificity. Ban persistent live scanning of the public. Ban use at protests, religious services, clinics, and political gatherings absent a court order tied to a serious threat. Require warrants for real-time deployment, exigent-circumstances reporting after emergencies, independent accuracy testing, retention limits, and automatic disclosure to defendants whenever facial recognition contributed to an investigation. Then allow retrospective use on lawfully obtained footage in serious felony cases, missing-person cases, and imminent threats, with corroboration required before any arrest. That framework is not “trust the algorithm.” It is “mistrust everyone, document everything, and still let detectives use lawful tools.” A very conservative sentence, frankly. And there is a practical governing reality here that shouldn’t be waved away with a dystopia reference and a tasteful shudder. If a city has video of a suspect committing repeated violent assaults on a subway platform, or footage connected to an abduction, refusing to run a tightly controlled facial-recognition query is not civil-libertarian elegance; it is self-imposed investigative blindness. Voters generally understand the distinction between a dragnet and a lead. The liberal concern about future abuse is legitimate, but the answer to a scalpel that can be misused is not banning surgery. It is limiting who can hold the scalpel, when they can use it, and what happens if they lie about where they put it. Less sci-fi sermon, more enforceable law.