Municipal affirmative ban on a class of AI use

A coalition runs a city- or county-level ordinance prohibiting a defined class of AI use — police facial recognition, predictive policing, public-housing surveillance, automated decision-making in benefits adjudication, school behavioural monitoring — before any specific deployment is on the table. The vehicle is local legislation passed by city council, county supervisors, or in some cases a citizen-initiated ballot measure, and the ordinance carries enforcement (private right of action, procurement bar, mandatory disclosure) rather than aspirational language. The instrument is built to be copied: model bills are published, sympathetic council members in other cities adopt the same template, and a patchwork accrues.

An actor chooses this strategy because the federal route on AI regulation is structurally slow and exposed to industry lobbying, while a city council is small enough that a coalition of a dozen community organisations can credibly carry the floor and a single sympathetic council-member sponsor can move the bill. Local bans also produce concrete operational facts on the ground — a police department actually loses the tool — rather than only changing future regulation, and the city-by-city patchwork eventually pressures vendors to standardise to the strictest jurisdiction or exit the market entirely.

It trades breadth for jurisdictional reach. A municipal ban protects a single city's residents and is preempted whenever a state legislature decides to centralise; vendors also route around bans by selling to neighbouring jurisdictions, to federal agencies operating in the city, or to private buyers the ordinance cannot reach. The accumulation is real but slow, and a coalition that wins ten city bans without state-level follow-through is still living one preemption bill away from a full reversal.

Verdict — wins the block, capped by the state

The strategy works on its narrowest claim — that a city council is a venue at which a coalition of community organisations can carry an affirmative prohibition of a class of AI use to enactment, against industry preference and police-department preference, faster than the federal or state route can produce a comparable instrument. The 17 United States jurisdictions counted by EFF in May 2022, the 26-jurisdiction CCOPS adoption arc, and the December 2020 Oakland extension into predictive policing, voice recognition, and gait recognition are the strategy's concrete record: real ordinances with teeth, in real cities, that real police departments lost real procurement access through. The BanFacialRecognition.com campaign and the Ban the Scan campaign supplied the national infrastructure — the interactive map, the model ordinance, the framing — that allowed each new city's coalition to argue from a documented trend rather than as a local anomaly. Oakland Privacy's standing Privacy Advisory Commission is the strategy's highest-coherence form: an ordinance that builds a permanent civilian oversight body, and the body then becomes the rail along which further prohibitions ride by amendment rather than re-litigation.

The strategy is good as a tool for converting principle into local operational fact, and bad as a path to closing the buyer market.

Strong on the operational claim. Seventeen US police departments actually lost facial-recognition procurement access; Oakland's police department lost predictive policing, voice recognition, and gait-recognition procurement access. Concert promoters at 40-plus major US festivals — Coachella, SXSW, Bonnaroo, Austin City Limits — committed to not deploying facial recognition in 2019 after Fight for the Future's artist-pressure track, and LiveNation and AEG, the two major US live-events conglomerates, publicly reversed posture in the same window. These are concrete operational facts the strategy produced that no other strategy in the corpus has produced at comparable density on the same harm. The pre-deployment framing — passing the ordinance before the police department has bought the system, so the political contest is over a future tool rather than a tool the department says it relies on — is the strategy's clearest tactical advantage over strategic litigation against algorithmic state decisions, which can only fire after a deployment has already produced an identifiable harm and an identifiable plaintiff.

Strong on the propagation claim. The model-ordinance form is the strategy's signature operational invention: a portable legislative artefact pre-engineered to reduce the per-city drafting cost from "write new law" to "adopt the template", and a portable coalition argument pre-engineered to make the political case in any council chamber rather than only in San Francisco's or Boston's. CCOPS is the corpus's clearest example of a model-ordinance campaign converting an organisational template into a multi-jurisdiction outcome. The model also feeds itself: every new city's enactment becomes evidence the next city's coalition cites, and the interactive map Fight for the Future built turns the patchwork into a propaganda artefact that recruits the next city.

Weaker on the closure claim. The strategy has produced no statewide full ban and no federal ban in seven years from the first San Francisco enactment, and the Facial Recognition and Biometric Technology Moratorium Act in the US Congress remains unpassed in 2026 despite being introduced in 2020 and re-introduced in 2023 with the BanFacialRecognition coalition as named endorsers. The 17-jurisdiction patchwork covers a small share of the US population, is geographically concentrated in progressive coastal cities, and leaves the populations most aggressively subjected to police AI tools — Black and Brown residents of un-banning jurisdictions, undocumented residents of jurisdictions where the federal partner (ICE, CBP) is the actual buyer, and travellers passing through TSA-controlled airports anywhere in the country — largely outside the strategy's protective shadow. The technology vendor is also not bound by the geography of any single jurisdiction: a vendor whose city contract is lost simply sells to the federal agency operating in the same city, to the neighbouring county, to private retailers, or to airports and venues the ordinance does not reach, and the strategy has not yet produced an answer to the routing-around problem.

Weakest on the durability claim. The 2022 New Orleans reversal of the 2020 ban — under political pressure tied to a specific violent-crime narrative — established that a municipal ordinance enacted without statutory durability is subject to rollback by the same council that enacted it, faster than the original political coalition can re-mobilise. The ACLU has itself documented the New Orleans Police continuing to use live facial recognition in violation of the local ordinance, which adds the further problem that even where the ordinance survives politically, enforcement against the police department whose practice it constrains depends on a council majority's appetite for that discipline at the moment a high-salience crime story breaks — and that appetite is rarely strong enough. The strategy can win the city; what it cannot yet do is hold the city it won.

The strategy is bad as the lone arm. A patchwork without a state or federal closure is a patchwork the technology grows around — through federal agencies, through neighbouring jurisdictions, through private buyers, and through state preemption — and the strategy's narrative win obscures that the unbanned majority of the buyer market continues to scale uninterrupted. The strategy needs the coalition lobbying of binding regional regulation arm in front of it at the state and federal level to convert municipal momentum into closure, and the strategic-litigation arm behind it to enforce the bans against the police departments that defy them. Run alone, the strategy produces a constellation of protected cities surrounded by a growing technology base; run as the local arm of a vertically-integrated campaign, it is the floor that makes the higher instruments politically possible.

Ecology

The strategy is fed by empirical audit and expose at the substrate. The Gender Shades-class audit findings on differential accuracy rates across race and gender are what made municipal bans politically possible: a council member who has the 34.7-per-cent-error-rate-on-darker-skinned-women number in hand has an argument the police department's procurement memo cannot answer, and the Algorithmic Justice League's research is cited in the legislative history of multiple US municipal facial-recognition bans. Without the audit number, the political argument collapses into a values dispute the police department wins by default.

It is paired with counter-narrative framing of AI harm as the public-facing register. The "ban-not-regulate" framing the Fight for the Future BanFacialRecognition.com campaign coined in 2019, and the "Ban the Scan" framing Amnesty International extended into a global campaign in 2021, are the strategy's working civil-society register — short, declarative, unhedged, and built to occupy the public-affairs ground a regulatory-compromise framing would otherwise take. Without the framing, the council debate defaults to "what safeguards should accompany the technology" rather than "should it be banned outright", and the council member loses the rhetorical ground before the procurement vote is taken.

It is fed by local rapid response against single deployment as the recruiting funnel. Oakland Privacy emerged from the 2013-2014 fight against Oakland's Domain Awareness Center — a single-deployment campaign — before it became the standing coalition that wrote the city's surveillance ordinance, won the Privacy Advisory Commission, and extended the prohibitions to predictive policing in 2020. The single-deployment fight builds the coalition the affirmative-ban strategy then deploys: a city without a recent or live single-deployment fight rarely has the political coalition ready to carry a model ordinance to enactment.

It feeds coalition lobbying of binding regional regulation as the precedent register for state and federal pushes. Every new municipal enactment becomes evidence the higher-level legislative coalition cites at the next legislative venue; the 17-jurisdiction count is itself a public-affairs artefact the Markey-Merkley-Jayapal-Pressley federal moratorium bill's endorsers use to argue that the country has already accepted the principle in a documented sample of cities and the federal floor is the next step. The strategy without the coalition-lobbying arm in front of it is a patchwork the technology grows around; the coalition-lobbying arm without the municipal precedent below it is an argument the state legislature has no evidence base to act on.

It is structurally adjacent to institutional procurement refusal of AI vendors — both strategies operate by denying a buyer market to the vendor rather than by regulating the technology's use — and the difference is which actor is doing the refusing. The municipal-affirmative-ban strategy uses the city government as the refusing actor; the institutional-procurement-refusal strategy uses universities, hospitals, school districts, libraries, or private firms. Together they map the universe of actors that can deny a specific buyer to an AI vendor without going through the state or federal legislative process at all, and the patchwork they jointly produce is the closest the movement has come to a market-side restraint on a deployment class.

It is structurally adjacent to humanitarian disarmament treaty in its logical form — prohibit a class of technology before deployment, by binding political instrument, at the level of the political community that holds standing — and the difference is scale. The humanitarian-disarmament strategy operates at the nation-state level through multilateral negotiation and reaches the global ceiling; the municipal-affirmative-ban strategy operates at the city level through council legislation and reaches the local floor. The two strategies face the same structural problem from the opposite direction: the multilateral instrument struggles to bind individual states, and the municipal instrument struggles to bind anything above the city limit. A movement that ran both arms in vertical integration — humanitarian-disarmament at the multilateral level, municipal-affirmative-ban at the city level, coalition-lobbying at the state and federal level between them — would have a layered prohibition-stack the AI-governance literature has not yet seriously prototyped.

It competes with strategic litigation against algorithmic state decisions for the same political coalition's time and attention on the same harm. The litigation arm rules after deployment on existing law against an identifiable plaintiff; the affirmative-ban arm rules before deployment on new law without a plaintiff. The two strategies are not substitutes — they backstop each other, the litigation arm enforcing the affirmative ban where police departments defy it and the affirmative ban setting the legal substrate the litigation arm rules against — but in any given jurisdiction the same civil-liberties organisations are the operative coalition for both, and the resource allocation between the two is itself a strategic question the movement has not closed. The La Quadrature du Net Technopolice campaign in France runs the litigation arm against municipal AI surveillance deployments in the absence of a French institutional vehicle for the affirmative-ban arm — a substitution that demonstrates which arm a movement runs is also a function of which legislative venue is available to it at all.

It is structurally upstream of open letter collective signatory action at the strategy-to-strategy level, even though the open letter is the older form. The municipal-affirmative-ban strategy uses open letters at every stage as launch and escalation vehicles — the 65-parliamentarians joint statement Big Brother Watch co-ordinated in October 2023, the Markey-Merkley-Jayapal-Pressley moratorium bill's endorser list, the #TireMeuRostoDaSuaMira open letter the Coding Rights-co-ordinated Brazilian #SaiDaMinhaCara campaign published in 2023 — and the open letter functions as the strategy's preferred coalition-public-record artefact at the moment a municipal coalition wants to demonstrate breadth.

The strongest competing strategy is state preemption — the industry-backed pattern of passing state legislation that overrides local surveillance restrictions. This is not a separate movement strategy with a separate adopter set in the corpus; it is the opposing political coalition's response to the municipal-affirmative-ban strategy, and the movement's structural disadvantage in the contest is that the state legislative venue is more exposed to industry lobbying than the city council is. The strategy's strongest hedge against preemption is not faster municipal accumulation but vertical integration with coalition lobbying of binding regional regulation at the state level, so the same coalition that holds the city ordinances is also fighting the preemption bill at the statehouse rather than letting the two fronts be decided by different actors at different paces.

Adopters seen so far

Beyond the canonical-side adopters linked above, the strategy has the densest adopter pattern of any strategy in the corpus at the local-group level. Oakland Privacy is the corpus's clearest example of a citizen-led local group whose principal mode is the affirmative-ban strategy applied amendment by amendment to a single jurisdiction's surveillance ordinance. The 13-state Brazilian #SaiDaMinhaCara campaign extends the strategy's adopter set out of the US-civil-liberties cluster and into a federated Latin American context where the political instrument is state and municipal bills rather than CCOPS-style model ordinances, demonstrating that the strategy's form is not specific to US municipal politics even where the operational template is. The geographic ceiling — the absence so far of the strategy's adoption at scale in continental Europe, South and Southeast Asia, Sub-Saharan Africa outside the Brazilian and South African cases — is the gap the strategy's next decade of growth is sitting against.