Faces Are Not Evidence: Ban Government Facial Recognition in Denver

The Problem

Denver runs a growing surveillance apparatus with almost no rules governing it. The city has no ordinance that restricts biometric surveillance, no public process before police acquire new tracking technology, and no prohibition on algorithmic tools that score residents by their predicted “risk.” Decisions about who gets watched, scanned, and tracked are made inside departments and vendor contracts - not in public, and not by the people being surveilled.

This is not hypothetical. Beginning with an 18-camera pilot in 2023, Denver built a network of 111 solar-powered automated license-plate-reader (ALPR) cameras across the city in 2024, manufactured by Flock Safety (Denverite, 2024). These cameras photograph every passing vehicle and log its location over time, building a searchable record of the movements of people who have done nothing wrong. There was no surveillance ordinance, no public vote, and no binding limit on who could search the data.

The consequences arrived quickly. Records obtained by the ACLU of Colorado and reported by NBC News showed that outside agencies queried Denver’s ALPR network more than 1,400 times for immigration-related purposes since 2024, including searches simply labeled “ICE” (NBC News, 2025). A surveillance tool sold to the public as a way to recover stolen cars became, in practice, an immigration-tracking database that Denver residents never consented to. The backlash was strong enough that the Denver City Council voted 12-0 against renewing the Flock contract in 2025, and the city moved to replace the system in 2026 (Denverite, 2026). But the underlying problem remains: there is still no law that requires this kind of debate before the next surveillance tool is purchased.

The Technologies Denver Has No Rules For

• Facial recognition. Denver has no ordinance banning or restricting city use of facial recognition. Colorado’s SB22-113 (2022) added warrant and disclosure requirements for state and local agencies, but it regulates use rather than prohibiting it, and it leaves municipalities free to deploy the technology within those limits.
• License-plate-reader networks. As the Flock episode showed, mass ALPR surveillance was deployed at scale with no ordinance controlling retention periods, sharing with federal agencies, or audit rights.
• Algorithmic risk-scoring and predictive policing. Software that assigns “risk scores” to individuals or forecasts where crime will occur can launder biased historical arrest data into automated suspicion, concentrating police attention on the same over-policed neighborhoods - with no Denver rule requiring it be tested, disclosed, or banned.

Why It Matters Who Bears the Error

Surveillance error is not distributed evenly. A landmark federal study of 189 facial recognition algorithms found that Asian and African American faces were falsely matched up to 100 times more often than white faces in one-to-one matching, with the highest false-positive rates among Native American, Asian, and Black demographics in U.S.-developed algorithms (NIST, 2019). These are not abstract error bars. In January 2020, Detroit police wrongfully arrested Robert Williams, a Black man, in front of his wife and two young daughters based on a false facial recognition match, detaining him for roughly 30 hours for a crime he did not commit (ACLU, 2021). His was the first publicly reported wrongful arrest caused by the technology - and the people most likely to be the next one look like the residents Denver’s policies are supposed to protect.

Our Solution

Denver should adopt a comprehensive surveillance and biometric privacy ordinance with five components.

1. Ban City Use of Facial Recognition

Prohibit any Denver agency - including the Denver Police Department and the Denver Sheriff Department - from acquiring, possessing, or using face recognition technology, or information derived from it. This includes real-time scanning, after-the-fact image matching, and indirectly obtaining results by asking another agency or a private vendor to run a search. This is the same approach taken by San Francisco in 2019 and Boston in 2020. A face is not a password residents can change after it leaks; banning the technology is the only reliable way to prevent the wrongful-arrest and demographic-bias harms it produces.

2. Restrict Civil-Use ALPR and License-Plate-Reader Networks

For any ALPR system the city operates or contracts for, the ordinance must require, by law and not by policy:

• Short, mandatory data deletion - non-hit plate reads deleted within days, not retained indefinitely.
• A prohibition on sharing ALPR data with federal immigration enforcement, consistent with Denver’s status as a city that does not use local resources for civil immigration enforcement.
• No commercial or out-of-jurisdiction data sharing without a specific, logged, individualized legal basis.
• Public audit logs - every search recorded with the searching agency, the user, and the stated reason, available to an oversight body.

3. Ban Algorithmic Risk-Scoring in Criminal Justice

Prohibit the city from using predictive-policing software or individual “risk scores” generated by automated systems to direct patrols, target individuals, or inform charging, bail, or sentencing recommendations. These systems train on historical arrest data that reflects decades of disproportionate enforcement, then recycle that bias as a neutral-sounding number.

4. Community Control Over Police Surveillance (CCOPS)

Adopt a community-control ordinance modeled on the laws now in force in 26-plus jurisdictions (ACLU, 2024). Before any Denver agency acquires or uses any new surveillance technology, it must:

• Publish a surveillance use policy describing the tool, the data it collects, retention, and sharing.
• Obtain City Council approval at a public hearing - the people, through their elected representatives, decide.
• File an annual surveillance report disclosing how each approved technology was actually used.

5. Enforcement With Teeth

The ordinance is only as strong as its remedies. Provide a private right of action allowing residents to sue for violations, suppression of evidence obtained through banned technology, and discipline for officials who circumvent the law - including by procuring banned results through a third party. Disclosure of any surveillance tool used against a criminal defendant must be made before trial.

Evidence

Cities That Have Already Acted

These bans have not produced lawless cities. They have produced a public process - one where the decision to deploy a powerful tracking technology is made openly, with the chance to say no.

The Technology Misidentifies People of Color

The most authoritative evidence on facial recognition’s accuracy comes from the federal government’s own testing laboratory. In its 2019 Face Recognition Vendor Test Part 3, the National Institute of Standards and Technology evaluated 189 algorithms from 99 developers and found systematic demographic differences: in one-to-one matching, Asian and African American faces produced false positives 10 to 100 times more often than white faces, depending on the algorithm, and one-to-many matching produced the highest false-positive rates for African American women (NIST, 2019). Notably, algorithms trained on more diverse data showed smaller gaps - confirming the bias is built in by design choices, not inevitable physics.

The real-world cost of that error is documented. Robert Williams’s 2020 wrongful arrest in Detroit - the first publicly known case of facial recognition causing a false arrest - led to an ACLU lawsuit and, ultimately, a settlement and policy changes for the Detroit Police Department (ACLU, 2021). The point of a ban is to ensure Denver never produces a Robert Williams of its own.

Local Context

Colorado has begun, tentatively, to regulate this space - but it has left the strongest protections to local governments. SB22-113 (2022) requires state and local agencies to file a notice of intent before using a facial recognition service, bars its use for ongoing surveillance or real-time identification absent a warrant or specific exceptions, prohibits its use to target people for their political, religious, or social views, and requires disclosure to criminal defendants (Colorado General Assembly, 2022). These are meaningful accountability measures - but they are not a ban, and they explicitly leave room for facial recognition to be used. Denver can and should go further within its home-rule authority.

The ALPR experience makes the case for local action concrete. Denver’s Flock network was justified to the public as an anti-auto-theft measure - and city officials credited it with vehicle recoveries and arrests after auto thefts fell from more than 12,000 in 2023 to roughly 8,550 the following year (NBC News, 2025). But the same network became a tool that outside agencies searched more than 1,400 times for immigration purposes, in a city whose policies are meant to keep local resources out of federal immigration enforcement (NBC News, 2025). When the contract came up for renewal, the City Council rejected it 12-0 (Denverite, 2026). That unanimous vote shows the political will exists. What is missing is a durable, technology-neutral ordinance so that the next surveillance tool faces the same scrutiny before it is deployed, not after a scandal. At the state level, SB26-070 (2026) sought to bar government access to historical location-tracking databases but stalled on Second Reading before the session adjourned, showing both that the surveillance-policy conversation is active in Colorado and that a durable local ordinance cannot wait on the legislature.

Frequently Asked Questions

“Doesn’t this tie the hands of police investigating serious crimes?” No. The ordinance bans facial recognition - a technology with documented, race-skewed error rates that has already produced wrongful arrests - and the unaccountable use of mass tracking tools. It does not stop police from using traditional investigative methods, responding to crimes, or obtaining a warrant for a specific search. It targets dragnet surveillance of everyone, not lawful investigation of someone.

“If the technology gets more accurate, why ban it instead of regulating it?” Even setting aside accuracy, perfectly accurate facial recognition would enable perfect, automated tracking of every resident’s movements and associations - a power no Denver agency should hold. And the bias is real today: the federal government’s own testing found error rates up to 100 times higher for Asian and Black faces (NIST, 2019). A ban prevents the harm; “regulation” of a tool this powerful tends to normalize it.

“Didn’t the license-plate cameras reduce auto theft?” City officials credited the cameras with vehicle recoveries and arrests, and auto thefts did fall after the pilot began (NBC News, 2025). But the same network was queried more than 1,400 times for immigration purposes, against the wishes of the community it surveilled. A safety tool that doubles as an immigration-tracking and location-history database is not a fair trade. The ordinance permits tightly controlled, audited, deletion-bound license-plate systems where the community approves them - it ends unaccountable ones.

“How is this different from what Colorado already passed?” SB22-113 (2022) regulates and discloses facial recognition use; it does not ban it, and it leaves the strongest choices to local governments (Colorado General Assembly, 2022). Denver’s ordinance would prohibit city facial recognition outright, add license-plate and predictive-policing rules the state law does not cover, and create a standing public-approval process for all future surveillance tools.

“Will this be expensive?” No. This is primarily a regulatory policy - the city saves money by not buying banned technologies and not renewing contracts like the lapsed Flock agreement. The only meaningful cost is modest staffing for the oversight and reporting process.

How We Pay For It

This policy is low-cost by design. It restricts and prohibits spending more than it requires it.

1. Net savings on technology contracts. Banning facial recognition and ending unaccountable ALPR contracts avoids procurement and subscription costs. The Flock contract that the City Council declined to renew was a six-figure annual expense.
2. Oversight and reporting staff. The primary new cost is modest: a small number of staff (within an existing office such as the City Attorney’s Office, the Auditor, or an independent monitor) to administer the use-policy review, maintain the public audit logs, and compile the annual surveillance report. This is comparable to the administrative cost of any new public-transparency mandate and is far smaller than the cost of a single wrongful-arrest lawsuit.
3. Avoided liability. Wrongful arrests driven by faulty surveillance technology generate litigation and settlements, as the Detroit case demonstrates (ACLU, 2021). Preventing those harms protects the city budget as well as residents.

References

• American Civil Liberties Union. (2021). Williams v. City of Detroit: Face recognition false arrest. https://www.aclu.org/cases/williams-v-city-of-detroit-face-recognition-false-arrest
• American Civil Liberties Union. (2024). Community control over police surveillance (CCOPS). https://www.aclu.org/community-control-over-police-surveillance
• Colorado General Assembly. (2022). SB22-113: Artificial intelligence facial recognition. https://leg.colorado.gov/bills/sb22-113
• Colorado General Assembly. (2026). SB26-070: Ban government access historical location information database. https://leg.colorado.gov/bills/sb26-070
• Grothaus, M. (2019, May 14). San Francisco passes city government ban on facial recognition tech. TechCrunch. https://techcrunch.com/2019/05/14/san-francisco-facial-recognition-ban/
• National Institute of Standards and Technology. (2019). Face recognition vendor test (FRVT) part 3: Demographic effects (NISTIR 8280). https://nvlpubs.nist.gov/nistpubs/ir/2019/NIST.IR.8280.pdf
• NBC News. (2025). Flock police cameras scan billions per month, sparking protests. https://www.nbcnews.com/tech/tech-news/flock-police-cameras-scan-billions-month-sparking-protests-rcna230037
• Denverite. (2026, February 24). Denver fires Flock, prepares to switch to new roadway surveillance system. https://denverite.com/2026/02/24/denver-ends-flock-contract-axon-alpr/
• WBUR News. (2020, June 23). Boston bans use of facial recognition technology. It’s the 2nd-largest city to do so. https://www.wbur.org/news/2020/06/23/boston-facial-recognition-ban