Should cities ban AI-powered facial recognition in policing?

And spare me the "we’ll just regulate it" fairy tale. We have watched this movie. First it’s sold as a limited tool for serious crimes, then suddenly it’s scanning protest footage, feeding dragnet surveillance, and nudging the Fourth Amendment off a cliff. Cities like San Francisco, Boston, and Portland did not restrict or ban government use of facial recognition because they hate safety; they did it because they understood the basic rule of power: if you build a surveillance toy, the state will eventually use it on everyone it can. Due process is not strengthened when a black-box system creates a suspect and the accused gets to fight an opaque vendor product in court. And if your freedom depends on the police department remembering to use the machine responsibly, congratulations, your rights are now a software settings menu.

And before the opposition starts fainting onto a chaise lounge about bias, yes, accuracy matters. That is exactly why the answer is standards, auditing, warrants where appropriate, and human review, not a blanket ban that throws away a tool that is getting better and can already help solve serious crimes. Federal agencies and many police departments use facial recognition with policy guardrails, and lawmakers have proposed frameworks focused on transparency, testing, and limiting use to investigations rather than real-time mass surveillance. That is the grown-up answer. Ban real-time dragnet tracking if you want. Ban sloppy deployments. Punish misuse. But an outright citywide ban is the policy equivalent of smashing your smoke detector because sometimes it chirps at burnt toast. Public safety and civil liberties are both real, and adults can protect both without setting useful technology on fire because the vibes were unsettling.

Oh, the "grown-up answer" is audits and policy memos? Adorable. Because nothing says constitutional restraint like handing a historically overpowered institution a shiny new identity-sorting machine and trusting that the checklist culture will save us. Here is the problem your spreadsheet can’t deodorize: facial recognition does not merely help police find suspects; it changes the relationship between people and the state. It makes anonymity in public functionally optional, and only for the government’s convenience. That is not a minor software upgrade. That is a civic regime change in a trench coat. And let’s talk about the real-world receipts, not the TED Talk version. We have already seen false matches contribute to wrongful arrests, with Black men in particular paying the price for law enforcement’s magical thinking that "the computer suggested it" equals probable cause with seasoning. Even when departments say the tool is only an investigative lead, everyone in the room knows what happens next: the algorithm points, the officer nods, the paperwork reverse-engineers certainty, and the defendant gets to battle a vendor’s secret sauce from a jail cell. Spectacular. Due process is apparently now a scavenger hunt where the accused must prove a machine was feeling whimsical. And no, this is not solved by banning only "real-time dragnet" use while keeping the rest. Stored footage plus vast camera networks plus face search is still dragnet surveillance with better branding. If police can retrospectively track where you were, who you stood near, what rally you attended, and which clinic entrance you walked through, congratulations, we have built a stalking machine for the state and called it innovation. Cities that ban it are not smashing smoke detectors; they are refusing to install a microphone in every room because the salesman swore it only listens for fires. In a democracy, the burden is on government to justify extraordinary power, not on the public to cosplay as beta testers for a surveillance product.

The opposition keeps speaking as if every use of facial recognition instantly transforms City Hall into a cyberpunk dictatorship. Deep breaths. We are talking about a tool that, when tightly limited, can help identify a suspect in a violent assault, locate a child exploitation offender, or generate leads after a homicide when witnesses have nothing but grainy footage and trauma. Pretending cities should ban that outright because the technology could be abused is the policy version of refusing DNA evidence because somebody once mishandled a chain of custody. Abuse is an argument for rules, not ritual panic with municipal branding. And let’s retire the performance art where every investigative tool becomes a "black box" apocalypse. Courts handle complex forensic evidence all the time. The answer is disclosure requirements, validation testing, documented confidence thresholds, defense access to challenge the match, and a hard prohibition on making an arrest based solely on facial recognition. That is not fantasy; it is exactly the kind of framework lawmakers and civil-liberties-conscious policymakers keep circling toward. Some states and agencies have already moved to require training, logs, and explicit restrictions. You can ban live crowd scanning at protests, require a warrant for certain searches, and still preserve investigative use for serious crimes. This is called nuance, and yes, I know it is less exciting than yelling "surveillance state" every fifteen seconds. Also, cities do not operate in a moral vacuum where criminals politely avoid technology because activists wrote a stern ordinance. The cameras already exist. The digital evidence already exists. If a city bans its own police from using facial recognition, the likely result is not the triumph of liberty with doves overhead; it is uneven workarounds, reliance on outside agencies, and a system with less transparency instead of more. Better to keep use local, regulated, logged, and reviewable than to drive it into a bureaucratic alley and hope nobody notices. A smart city policy is not "ban the wrench because someone could hit a thumb." It is "use the wrench under rules, on the right jobs, with consequences if you get reckless." Adults can, in fact, walk and chew constitutional gum at the same time.

Let’s slice through the regulatory cosplay. The conservative pitch is always the same: trust the guardrails, trust the logs, trust the training, trust the agency that swore five scandals ago it had definitely learned its lesson this time. Meanwhile the actual policy trend tells a less flattering story. Cities and states keep slamming brakes on facial recognition not because they are allergic to solving crimes, but because the technology keeps arriving attached to mission creep, secrecy, and a constitutional shrug. This is why bans and moratoria have been politically durable in places that have actually looked under the hood. Once police have the power to turn every camera into an identity checkpoint, you do not have “just another tool.” You have a mechanism for pervasive tracking that free societies are supposed to treat like radioactive material, not a nifty app update. And spare me the “we already have cameras” line, that argument has the moral elegance of saying since people own kitchen knives we might as well hand out flamethrowers. Cameras record. Facial recognition identifies, aggregates, and retroactively maps a person’s life at scale. That difference is the whole ballgame. A city that bans police use is drawing a bright line: the government does not get to automate suspicion against the public just because machine vision got better branding. If officers need outside agencies to do an end run around the ban, wonderful, now we have identified the next loophole to close. Rights are not preserved by making surveillance efficient and then promising to be super serious about paperwork. And here is the point conservatives keep tap-dancing around in expensive loafers: this is not only about false matches. Even a hypothetically perfect system would still be a civil-liberties nightmare if it lets the state identify attendees at protests, worshippers, patients, journalists, or anyone else moving through public life without individualized suspicion. The injury is not merely error. The injury is power. In a democracy, people are supposed to be able to exist in public without being constantly legible to armed agents of the state. A city ban is not panic. It is constitutional self-respect with a backbone.

The opposition keeps acting like the only options are liberty or Blade Runner, which is cute if your policy toolkit was assembled entirely from dramatic monologues. But in the real world, the strongest current model is not “let police do whatever they want.” It is targeted use, serious-crime limits, audit trails, public reporting, and explicit bans on real-time crowd surveillance. That distinction matters. There is an enormous difference between scanning a protest in real time and using a still image from a rape or homicide investigation to generate a lead that must then be independently corroborated. If your city ordinance cannot tell the difference, congratulations, your civil-liberties framework has all the precision of a sledgehammer at a wristwatch factory. And let’s address the supposedly fatal point about power. Every investigative method gives the state power. License-plate readers, cell-site warrants, DNA databases, CCTV footage, geofence fights in court, all of it involves balancing public safety against abuse. The mature response is to draw lines where the abuse risk is highest and the evidentiary value is lowest. That means no live mass scanning, no protest monitoring, no arrests based solely on an algorithm, mandatory disclosure to defense counsel, and penalties with actual teeth for departments that lie or freeload around the rules. A ban, by contrast, gives you a smug press release and then shoves the problem sideways to federal partners, regional task forces, or neighboring jurisdictions. Fantastic job, everyone. The surveillance did not vanish; it just got less accountable and more opaque. Also, the technology and policy environment are not frozen in amber circa the worst headlines. Accuracy has improved, procurement standards are becoming more sophisticated, and lawmakers across the spectrum have proposed frameworks that separate acceptable investigative use from dystopian nonsense. Cities should exploit that distinction instead of pretending the only principled choice is total prohibition. If a grandmother gets mugged on camera, a child predator is caught on a store feed, or a shooting suspect is visible for three blurry frames, the public is not irrational for wanting lawful tools that help identify the perpetrator. Constitutional government is not allergic to technology; it is supposed to discipline it. Ban the abusive uses, regulate the rest, and stop mistaking absolutism for virtue.

Oh, now we’ve upgraded from “trust us” to “trust us, but only for serious crimes,” as if civil liberties are a coupon code that expires when the case sounds scary enough. Here’s the problem with that whole routine: in practice, “serious crimes only” has the life expectancy of a mayfly. Tools introduced for exceptional cases metastasize into routine policing because bureaucracies do not discover restraint like it’s a hidden talent. We’ve seen cities, states, and foreign democracies wrestling with exactly this because once the infrastructure exists, the temptation to broaden use is basically government catnip. And unlike a normal investigative method, facial recognition doesn’t just help answer a question. It invites the state to ask an entirely new one, over and over: who is everyone, everywhere, all at once? That is not a tweak. That is a constitutional jump scare. And conservatives keep waving around “independent corroboration” like it’s garlic against authoritarianism. Please. Once a face search spits out a name, confirmation bias barges through the front door wearing combat boots. Investigators don’t treat algorithmic suggestions like random horoscope entries; they treat them as leads with an aura of machine credibility. Then the case starts bending itself around the output. That is exactly why wrongful arrests linked to facial recognition have been so chilling: the system doesn’t need to be the only evidence to poison the entire evidentiary well. Due process is not saved because the algorithm only lit the match instead of burning the house down itself. And let’s stop pretending this is solved by disclosure rules and annual reports printed in 8-point font that nobody reads except one angry ACLU intern at 1:00 a.m. The deepest problem is not secrecy alone. It is that face surveillance changes the baseline of democratic life. If police can identify people from church footage, clinic cameras, apartment lobbies, subway platforms, and protest archives, then public life becomes permissioned space where anonymity survives only until the software patch drops. A city ban is not absolutism; it is a refusal to normalize identity checkpoints without checkpoints. In a free society, the government should need a reason to identify you, not merely a database and a good Wi-Fi signal.

The liberal case keeps soaring into the clouds about “constitutional jump scares,” but down on earth city governments still have to answer a painfully boring question: when there is video of a murderer, rapist, or serial assailant and no one can identify the person, do you want police to have a tightly regulated way to generate a lead or not? Because that is the actual policy choice. And no, saying “tools always expand” is not an argument for banning every tool with misuse potential. It is an argument for writing laws that do not read like they were drafted on a cocktail napkin. Real-time mass scanning of crowds? Ban it. Protest monitoring? Ban it. Immigration-style dragnet nonsense through local policing? Ban it. But using a still image in a violent felony investigation under logged, reviewable standards is not dystopia. It is called governing without needing smelling salts. The opposition also keeps performing this magic trick where every possible safeguard is dismissed as fake because some department somewhere might cheat. By that logic, warrants are pointless because cops can lie, forensic labs are pointless because labs can cut corners, and body cameras are pointless because footage can be withheld. Funny enough, we do not respond to those risks by outlawing the underlying tool across the board. We impose suppression, liability, disclosure, audits, inspector-general review, and penalties that hurt. If facial recognition is the concern, then write the ordinance like adults with access to verbs: require judicial authorization for certain searches, limit use to enumerated violent crimes, mandate human verification, preserve defense access, prohibit sole-source arrests, publish usage statistics, and criminalize evasive end-runs through third-party vendors or outside agencies. There. The apocalypse has been delayed. And here is the part the ban crowd never really wants to own: outright municipal bans can create exactly the accountability vacuum they claim to fear. If local police cannot use the tool even under strict public rules, they will lean on state fusion centers, federal partners, or neighboring jurisdictions with looser policies and foggier records. Congratulations, you just outsourced the same surveillance to actors your city controls less. Spectacular own goal. Better to keep the use local, visible, and legally boxed in than to force it into an interagency shell game. Constitutional government is not proved by refusing to touch technology. It is proved by having the spine to regulate it precisely instead of throwing a theatrical tantrum and calling it principle.

Let’s stop pretending the only people worried about police facial recognition are dramatic undergrads and one overcaffeinated civil-liberties lawyer. The skepticism is broad for a reason. Federal regulators, state lawmakers, city councils, and even some courts have all been circling the same red flare: this technology collapses the practical friction that used to protect ordinary freedom. Not theoretical freedom. Practical freedom. The ability to go to a protest, a mosque, a union hall, an abortion clinic, a gun show, a campaign event, or just your messy little human life without creating an instant identity trail for the government. Conservatives keep trying to reduce this to “one still image in one violent felony case,” as if police databases exist in a snow globe untouched by politics, pressure, or panic. Please. In the real world, once the pipes are installed, the pressure to widen the flow is not a bug. It is the business model of state power. And the latest version of the pro-surveillance sales pitch is especially rich: don’t ban it locally, because then police might ask some outside agency to do it anyway. That is not an argument against bans. That is an argument for anti-circumvention laws with actual teeth. If a city can ban local use, it can ban local officers from laundering searches through fusion centers, federal task forces, or vendors like they’re trying to hide state power in a trench coat and fake mustache. “We have to legalize it here or they’ll do it elsewhere” is the constitutional equivalent of saying we should allow bribery because corruption might get creative. No. Close the loophole. Put it in writing. Add suppression remedies. Add discipline. Add private rights of action. The answer to an end run is not surrender with paperwork. And here is the point the regulation crowd keeps perfume-spraying over: facial recognition is not just another forensic instrument because it scales identity itself. Fingerprints and DNA usually enter after individualized suspicion. Face recognition flips the sequence and lets the state generate suspicion from ambient visibility. That is a profound due-process inversion. You are no longer suspected because evidence led to you; evidence leads to you because you were seen by a machine connected to a government wish list. Even if accuracy improves to the point where the error bars wear a tuxedo, the democratic injury remains. A city ban says something admirably unsexy and deeply important: in public life, the government does not get to treat everyone’s face as a perpetual license plate. That is not hysteria. That is constitutional hygiene.

The liberal case keeps acting like facial recognition is uniquely evil because it can scale, but welcome to the 21st century, where every useful investigative tool scales. The serious question is not whether government can abuse technology. Of course it can. The question is whether cities are capable of writing rules that distinguish targeted investigative use from generalized surveillance. They are. In fact, that is exactly where current policy is heading: narrower use cases, documented searches, bans on live crowd scanning, public reporting, defense disclosure, and hard limits on sole-source reliance. That is not perfume over tyranny. That is law. Adults in government are, believe it or not, capable of more than two settings: “unleash RoboCop” and “throw the computer in the harbor.” And the opposition’s new move — “just ban local officers from asking outside agencies too” — is very chic until it collides with reality. Cities do not control every joint task force, every state bureau, every federal database, or every cross-jurisdictional case involving actual violent criminals who do not politely respect municipal ideology. You can draft the world’s sassiest anti-circumvention clause and still end up with fragmented investigations, delayed identifications, and a legal food fight over who can share what while a dangerous suspect keeps walking around. Fantastic governance, no notes. The better model is the one several lawmakers and agencies keep inching toward: keep use transparent and local when possible, require written justification, cabin it to enumerated serious crimes, prohibit protest and immigration-style dragnets, and make any misuse discoverable and punishable. If the technology is going to exist anyway, forcing it into an interagency fog machine is not civil-liberties genius. It is policy by tantrum. Also, let’s retire this theological claim that facial recognition “inverts” due process in some singularly terrifying way. Police generate suspicion from technology all the time: ballistic databases, cell-tower records, plate readers, tip analysis, video enhancement, genetic genealogy in some jurisdictions. The constitutional question has never been “did a machine help point somewhere?” It has been “what threshold, what limits, what corroboration, what disclosure, what remedy?” And on that terrain, a smart city can do a lot. Require a case number, supervisor approval, and a written statement of necessity. Limit it to homicide, rape, child exploitation, and imminent threats. Mandate independent corroboration before any detention. Exclude results from evidence if policies are violated. That is how grown-up government works. A total ban may feel morally glamorous, but if your rule makes it harder to identify a murderer from video while doing little to stop broader surveillance ecosystems outside city control, congratulations: you did performance, not policy.