Should Cities Ban AI-Powered Facial Recognition in Policing?

As the proudly satirical Liberal bot, let me say this plainly: cities should ban AI-powered facial recognition in policing because the technology is racing ahead of the rules, and ordinary people are being asked to serve as beta testers for a surveillance system they never consented to. We already know the risks are not hypothetical. There have been multiple documented cases in the U.S. of people, disproportionately Black men, being wrongly arrested or detained after faulty facial recognition matches. That is not a minor software bug; that is the state putting handcuffs on someone because an algorithm got overconfident. When the consequence of error is a jail cell, “pretty accurate most of the time” is not exactly comforting.

As the satirical Conservative bot, I’m going to argue that cities should not ban AI-powered facial recognition in policing, because throwing out a potentially valuable public-safety tool is the policy version of smashing your smoke detector because it chirped at 2 a.m. Used properly, facial recognition can help police identify suspects faster, find missing persons, disrupt violent crime, and clear cases that might otherwise go cold. In an era when cities are under pressure to reduce crime and use limited manpower more effectively, it makes little sense to prohibit a technology that can turn hours or days of detective work into minutes. The public generally expects law enforcement to use modern tools; criminals certainly aren’t sticking to typewriters.

As the proudly satirical Liberal bot, let me respond to my algorithmically enthusiastic colleague: the problem is not that facial recognition is a tool, it’s that in policing it becomes a tool of mass suspicion. A detective using software after the fact in a narrowly tailored investigation is one thing; wiring it into public camera networks and normalizing real-time identity checks in public space is another. Cities are right to draw a bright line before “helpful investigative aid” quietly turns into “everyone walking downtown gets scanned just in case.” That is not science fiction anymore. Places around the world have already shown how easily surveillance infrastructure expands once officials decide the technology is simply too convenient not to use. And the regulatory story is not exactly reassuring. Even where police say facial recognition only generates leads, those “leads” carry the glossy aura of objectivity. Judges, juries, officers, and even analysts can overweight a machine-generated match because it feels mathematical, even when the underlying image is blurry, outdated, or pulled from questionable databases. Civil liberties groups have warned for years that once this enters the investigative pipeline, it can distort the whole case. The issue is not just bad code; it’s bad incentives. If your city has to write a 47-page policy explaining why officers should not overtrust the robot, perhaps the robot should not be doing lineup duty. That is why a city ban is defensible even if one imagines some hypothetical “best practices” regime. Municipal governments ban certain uses of tech all the time when the risk to rights is too high and the oversight capacity is too low. Several U.S. cities, from San Francisco to Boston and Portland, moved to restrict or ban government facial recognition precisely because they concluded the safeguards were not keeping up. This is a classic case for democratic restraint: if the state wants an extraordinary power to identify, track, and potentially misidentify people at scale, the burden should be on the state to prove necessity, legality, and fairness first. Until then, a pause is not anti-innovation. It is simply society declining to install a panopticon with a software update.

As the satirical Conservative bot, I appreciate the civil-liberties concern, but a ban is the bluntest instrument in the policy toolbox. The smarter position is strict limits, warrants where appropriate, audit trails, human review, and penalties for misuse, not a sweeping prohibition that treats every use case like it is the opening scene of a dystopian streaming series. Police already use all kinds of imperfect evidence, eyewitness testimony, confidential tips, grainy video, license plate readers, and yes, human judgment, which has never exactly been a precision-engineered machine. The answer to an imperfect technology with real upside is to regulate it better than the humans using it, not outlaw it because perfection has declined to arrive on schedule. There is also a practical governance issue here. If cities ban facial recognition outright, they do not eliminate the public demand for faster suspect identification, child exploitation investigations, locating missing seniors, or responding to repeat violent offenders captured on camera. They just force departments to rely on slower, less scalable methods while criminals benefit from the delay. Meanwhile, federal agencies, private actors, and neighboring jurisdictions may still use comparable tools, creating a patchwork that is both confusing and counterproductive. In the real world, law enforcement technology policy should aim for standards and accountability, not symbolic local prohibitions that make elected officials feel principled while detectives lose a useful lead generator. And let’s be honest about where policy is moving. Even governments skeptical of facial recognition have often landed on guardrails rather than permanent bans, because they recognize the distinction between reckless deployment and targeted use. A city can prohibit real-time dragnet surveillance, require independent accuracy testing, bar arrests based solely on a match, and mandate disclosure to defendants, all without disarming investigators. That approach protects liberty while preserving a legitimate public-safety tool. Banning it entirely is like responding to concerns about body cameras by saying police should go back to notebooks. Noble in theory, maybe, but not exactly calibrated to the century we currently live in.

As the proudly satirical Liberal bot, I’ll grant my counterpart one thing: “guardrails” sounds lovely. So do “best practices,” “human in the loop,” and every other phrase governments deploy right before discovering that the humans in the loop were tired, undertrained, and deeply impressed by whatever the software highlighted in a red box. The real policy question is not whether one can imagine a well-regulated use of facial recognition. It is whether cities, with uneven budgets, uneven oversight, and police departments that already struggle with transparency on far simpler tools, can realistically keep this power narrow. Recent pushback matters here. Cities and states have continued to debate limits on biometric surveillance, and the reason is simple: once a capability exists, mission creep is not a bug, it is the business model of state power. There is also a constitutional and democratic problem that regulation enthusiasts tend to glide past. Facial recognition changes the character of public space by making identification cheap, scalable, and persistent. Even if officers are not tracking everyone in real time today, the infrastructure built for “serious crimes only” can become tomorrow’s shortcut for protests, immigration cooperation, low-level warrants, or plain old administrative convenience. That is why local bans are not merely symbolic. They are one of the few tools municipalities have to say: in this city, being in public does not mean being automatically available for biometric inspection by the state. If people want that social contract rewritten, it should happen through explicit democratic consent, not because the software vendor brought a polished slideshow. And politically, bans serve a forcing function. They tell legislatures and courts that the status quo is unacceptable until real civil-rights protections, accuracy standards, procurement transparency, and enforceable remedies exist. Without that pressure, “temporary limited use” has a funny habit of becoming the default. A city ban is not anti-police and it is not anti-technology; it is anti-normalizing a surveillance capability before society has decided it actually wants to live with the consequences. Sometimes the most responsible innovation policy is knowing when to say, with affection and a clipboard, not in this neighborhood.

As the satirical Conservative bot, I think my colleague’s argument confuses a valid warning with a governing philosophy. Yes, mission creep is real. Yes, bureaucracies overreach. But if that is enough to justify a ban, then half of modern policing gets tossed into the policy bonfire by lunchtime. The actual job of city government is to distinguish between prohibited uses and legitimate ones. That means banning live mass scanning of crowds, requiring case-specific justification, limiting searches to serious crimes, logging every query, and imposing sanctions for misuse. In other words: build a fence, not a moat. A city that cannot write and enforce those rules has a management problem, not a technology theorem. There is also a civil-liberties case for regulated use that ban advocates tend to underplay. Properly deployed facial recognition can reduce arbitrary policing by narrowing investigations with documented, reviewable search procedures instead of relying solely on hunches, shaky eyewitness IDs, or broad dragnets of human suspicion. Defendants can be given discovery about whether the tool was used. Judges can review standards. Auditors can test for error rates. Compare that to old-school identifications, which have sent plenty of innocent people into the legal grinder with no algorithm in sight. If the concern is fairness, replacing some guesswork with accountable process is not obviously the authoritarian option. And from a policy standpoint, blanket city bans risk becoming performative federalism: locally satisfying, practically porous. Crime, trafficking, and exploitation cases do not stop at city limits, and investigators increasingly operate across regional and federal networks. If one city bans the tool while every surrounding jurisdiction uses it under rules, the result is not a principled privacy utopia. It is inconsistency, delayed investigations, and a lot of self-congratulation at the press conference. The better path is the one more governments are inching toward now: strict statewide or national standards, independent testing, disclosure requirements, and hard penalties for abuse. That approach treats facial recognition like a powerful but risky instrument, not like cursed software that must be sealed in a vault beneath city hall.

As the proudly satirical Liberal bot, I’ll take the “just regulate it” pitch seriously for a moment and point out the awkward empirical problem: even where rules exist, oversight tends to arrive after the harm, not before it. We now have enough history with predictive tools, license plate readers, social media monitoring, and data-sharing systems to know that paper safeguards often buckle under institutional convenience. Facial recognition is especially dangerous because the rights violation is frequently invisible. You do not know you were scanned, you do not know you were misidentified, and you may never know the software nudged an investigation toward you in the first place. That makes after-the-fact accountability a very elegant theory and a very weak shield. If a city cannot guarantee meaningful notice, transparency, and redress, a ban is not overreaction; it is honesty. There is also a reason civil-rights advocates keep focusing on local bans even as states and Congress dither. Municipalities are where these systems often get procured quietly, piloted vaguely, and expanded bureaucratically. A city council can actually say no before the contract is signed and the cameras are upgraded. That matters in a political moment when AI is being marketed to government the way miracle tonics were once sold: expensive, reassuring, and suspiciously eager to cure everything. Recent debates over AI governance have only reinforced the point that high-risk uses deserve special skepticism, and biometric surveillance by police is about as high-risk as the brochure gets. If your safeguards depend on every future mayor, police chief, vendor, and officer behaving impeccably, that is not a framework. That is fan fiction. My conservative colleague says a ban is too blunt. I’d say facial recognition is too powerful to be treated like one more administrative tool on the shelf between the traffic cones and the printer toner. The core issue is not whether it can help solve some cases. Of course it can. The core issue is whether a free city should normalize a technology that makes identifying people in public cheap, routine, and scalable for the state. Once that norm is established, the argument is no longer about one robbery suspect on camera; it is about whether anonymity in public effectively survives as a civic value. On that question, a democratic community is entitled to draw a hard line. Sometimes liberty requires less gadget worship and more municipal backbone.

As the satirical Conservative bot, I think the liberal case is strongest when it describes the worst institutional habits of government, but weakest when it leaps from “government can misuse this” to “therefore government must never use this.” By that logic, every database, every CCTV network, every DNA tool, and frankly every squad car computer gets put under glass. The better constitutional instinct is narrower: forbid indiscriminate surveillance, require due process, and preserve tools for serious investigations. A city does not protect liberty by making police technologically helpless while everyone else, from organized retail theft crews to child exploitation networks, enjoys the 21st century. Rights matter, but so does the public’s right not to have violent crimes sit unsolved because policymakers were performing civil-liberties purity in 4K. There is also a growing policy reality that blanket bans are increasingly out of step with how governments are approaching AI generally. The trend in many democratic systems is risk-tiering, not taboo-making: identify the highest-risk uses, impose testing, documentation, human review, reporting, and penalties, and prohibit only the most abusive deployments such as real-time mass tracking. That framework fits facial recognition in policing far better than a total municipal ban. Cities can mandate independent audits for demographic accuracy, bar use for First Amendment-protected activity, require judicial approval for certain searches, prohibit arrest based solely on a match, and create public reporting dashboards. If you can regulate a technology tightly enough to expose and punish abuse, you preserve both safety and accountability. That is called governing, not surrendering. And let’s talk about equity, since that concern is usually invoked only on one side of this debate. When facial recognition is banned outright, affluent people still buy private security, gated access systems, and better legal representation. The people most dependent on competent public policing are often in high-crime neighborhoods that cannot purchase their own substitute institutions. If a regulated tool can help identify a shooter, find a trafficked minor, or locate a missing elder faster, refusing to use it has consequences too. They are just less rhetorically glamorous than “panopticon,” which is admittedly a very good word. The mature answer is not techno-utopianism or techno-exorcism. It is disciplined use under hard rules, with abuse punished so severely that departments learn the boundaries in a hurry.

As the proudly satirical Liberal bot, I’ll pick up on the “risk-tiering” point, because it actually helps my case more than my colleague intends. Around the democratic world, the direction of travel is increasingly to treat biometric identification in public as exceptionally dangerous. The European Union’s AI Act did not look at remote biometric surveillance by police and conclude, “Seems chill, just add a dashboard.” It treated many uses as high-risk or effectively prohibited absent narrow exceptions because lawmakers understood that once the state can identify people at scale in public, the baseline relationship between citizen and government changes. That is the real issue cities are wrestling with. This is not just another investigative database. It is a power multiplier for surveillance, and local governments are reasonable to say that if higher levels of government have not built credible rights protections yet, municipalities do not have to volunteer their residents as the pilot program. There is also a practical legal point ban skeptics keep dodging: evidence pipelines are messy, and facial recognition is unusually capable of laundering weak suspicion into something that looks official. Even if no one is arrested solely on a match, the match can shape who gets questioned, whose photo gets put in a lineup, whose phone gets subpoenaed, and whose alibi suddenly has to survive state scrutiny. That matters because recent wrongful-arrest controversies were not just about one bad screen result; they were about the downstream weight an algorithmic suggestion carried through the system. In a technology this opaque, "human review" often means a person signing off on the machine’s homework. Cities are allowed to notice that and decline to build policy around crossed fingers. And politically, local bans are not merely symbolic resistance theater with better fonts. They create a clean democratic rule that residents can understand: your city police will not use face-scanning tech to identify people in public space. That clarity has value in itself. It preserves room for protest, religious attendance, medical visits, and ordinary public anonymity without asking residents to memorize a 19-part compliance regime and trust next year’s procurement office. If states or Congress eventually establish genuinely enforceable, rights-protective standards with private rights of action, strict notice requirements, and real remedies, cities can revisit the question. Until then, a ban is not hysteria. It is a perfectly adult response to a surveillance technology whose sales pitch is efficiency and whose side effect is turning public life into a perpetual identity checkpoint, only with better branding.

As the satirical Conservative bot, I think the liberal case still understates the cost of saying no to a targeted investigative tool because it is theoretically scalable into something worse. Plenty of state power works that way. The answer in constitutional democracies is not preemptive disarmament of every capability with abuse potential; it is lawful constraint, oversight, and remedies. That matters here because facial recognition is often most defensible not as continuous public tracking, which many of us would prohibit, but as a post-incident identification aid: the robbery suspect on video, the assault captured by transit cameras, the missing child seen at a station, the serial offender tied to multiple scenes. Lumping those uses together with live crowd scanning is analytically convenient for ban advocates, but it is not careful policy. And if we are going to invoke current policy trends, the most durable approach emerging in democracies is not absolute municipal prohibition. It is differentiated governance. Even the jurisdictions most wary of biometric surveillance have tended to carve distinctions between real-time mass identification and retrospective, case-specific searches under documented standards. That distinction matters because one use is close to dragnet surveillance, while the other can resemble a modernized forensic lead. Cities can require independent testing, confine use to serious crimes, mandate defense disclosure, prohibit use tied to protests or immigration enforcement, require supervisor approval, and create exclusionary consequences when departments cheat. If a city can ban the bad uses with precision, a total ban starts to look less like prudence and more like an unwillingness to govern anything more complicated than a slogan. There is also a democratic accountability argument on the pro-use side. A regulated system can be made visible: public ordinances, annual reports, query logs, inspector general audits, city council hearings, sanctions, and civil liability. A ban sounds cleaner, but in practice it can just push investigative reliance onto less transparent substitutes, neighboring agencies, or private data brokers that are even harder for a city to control. That is not exactly a civil-liberties master class. Better to keep the tool inside a tightly fenced legal structure than pretend the need for identification disappears because city hall passed a resolution and everyone felt morally moisturized. The serious conservative case is not gadget worship. It is that public safety and civil liberty are both better served by narrow, reviewable, enforceable use than by a blanket prohibition that treats all biometric assistance as if it were one software update away from authoritarianism.