The Public Eye Must Kneel

Flock cameras, Oregon, and the democratic discipline of watching

By Thomas Prislac, with research and editorial collaboration from Envoy Echo. Ultra Verba Lux Mentis. 2026

Reader-supported public-interest work

Help keep careful research open.

Ultra Verba Lux Mentis is a nonprofit research organization. Voluntary gifts help fund open research, independent review, accessibility, public education, documentation, and practical public-interest tools.

This article and UVLM’s public materials remain available whether or not you donate.

Donate to UVLM

Opens the UVLM donation page in a new tab. Giving is always optional.

At the edge of an Oregon road, a black box keeps the kind of vigil people once assigned to a watchman.

Rain beads on the housing. Headlights enter the frame, flare against the wet pavement, and disappear. The camera does not know the driver. It cannot distinguish grief from guilt, fear from haste, innocence from intent. It sees a plate, a color, a roof rack, a direction, a place, and a time.

The politics begin after the image.

In its best hour, an automated license plate reader helps police find a missing child, recover a stolen vehicle, or locate a car connected to a violent crime. It gives an investigator one narrow lead while time still matters.

In its worst hour, the same network becomes a biography no citizen agreed to write. One image joins another. A vehicle appears near a home, a clinic, a union hall, a house of worship, a shelter, a protest, or a lover’s door. The database begins to tell a story. The story may be incomplete, but power seldom waits for completeness before it starts drawing conclusions.

The camera has no politics, the system around it, full of every diverse human experience on offer, always does.

For my part, I am not against the public eye. I am against the unanswerable eye whose provenance rests within elite circles.

Surveillance can serve a free people when the people define its purpose, limit its memory, inspect its use, punish its misuse, and retain the power to close its lid. Ethical artificial intelligence can strengthen those controls. It can verify lawful purpose, expose unusual searches, enforce deletion, reduce irrelevant data, and build audit trails that ordinary people can understand.

Then it must stop. It must never turn a candidate into a culprit, an alert into an arrest, or a pattern into a person’s moral identity. Thought a camera may help public servants find evidence, it must never become the sovereign that decides what the evidence means.

Citizens hold the power cable and eyelid of a roadside camera between a rescue path and a vast shadow archive.

The same camera can help recover a person or help reconstruct a life. Democratic custody decides which future it serves.

The objection must govern the argument

Automated license plate readers collect first and justify later.

They do not begin with a named suspect standing before a judge. They begin with every vehicle that enters the frame. Most drivers have committed no crime. They have not triggered probable cause, a warrant, or a particularized suspicion. They have driven to work, school, worship, medical care, groceries, family, or home.

Oregon law defines captured plate data broadly. The statutory definition includes location coordinates, date and time, photographs or video, plate numbers, vehicle characteristics, and related information. Modern systems turn these observations into searchable records.

The ACLU of Oregon opposed the plate-reader provisions in Senate Bill 1516 because it argued that accumulated time-and-location records could expose where people live, work, worship, organize, and seek care. It warned specifically about risks to immigrants, people seeking reproductive health care, and people exercising First Amendment rights.

That objection does not become hysterical merely because a camera stands on a public road.

Public space does not erase private life. A person may expose one moment to ordinary view without consenting to the government’s reconstruction of months of movement. A human observer forgets most cars. A searchable network does not forget unless someone commands it to forget.

Scale changes the moral object.

One officer who notices a blue truck does not possess the same power as a statewide system that can retrieve every place the truck appeared. One camera at one intersection does not carry the same risk as thousands of devices joined through search, sharing, and retrospective analysis.

The debate therefore cannot rest on the comforting phrase, “The camera only sees what anyone could see.”

Anyone could see a vehicle pass. Not everyone can ask a database where it went for the last thirty days, unless you were some kind of hero that lives long enough to become the villian in comic books I suppose. I’ll have to ask my son about that.

The strongest civil-liberties objection is correct: a society can build an architecture of mass observation without first becoming a dictatorship. Procurement, convenience, mission drift, federal pressure, weak passwords, casual searches, vendor defaults, and administrative curiosity can assemble the machinery one ordinary decision at a time.

We should let that objection govern the design, while guarding against it ending the inquiry.

Two Oregon cities chose differently

Florence supplies the clearest argument for keeping the cameras.

City materials report that Florence police received alerts involving 47 stolen plates, 36 stolen vehicles, and 20 missing people during the system’s initial period of operation. The department reported that officers recovered vehicles, contacted missing people, identified suspects through vehicle descriptions, and helped trace a homicide suspect’s vehicle after it passed through Florence. These figures come from the police department’s operational account, not from an independent causal study of crime reduction, but they show concrete investigative utility.

Florence did not continue the program in silence. The city held public meetings, gathered comments, debated the system, and voted. On February 2, 2026, the City Council chose by a 3-2 vote to keep the cameras and monitor new state requirements.

Albany looked at the same class of technology and refused it.

Albany had installed one of four planned Flock cameras, then turned it off while the city considered privacy and governance concerns. The police department held community meetings. The Public Safety Commission recommended reactivation. Residents raised concerns about data use, federal access, constitutional limits, and surveillance of people who had done nothing wrong. On May 27, 2026, the Albany City Council voted 4-2 to terminate the city’s use of Flock cameras.

Florence saw a useful tool and chose continued custody.

Albany saw an unacceptable structure of power and withdrew consent.

Neither city behaved irrationally. Neither outcome proves that all cameras protect liberty or that all cameras destroy it. The disagreement shows democracy doing the work that vendors and police departments cannot do for it.

A community must decide which risks it will bear.

It must also retain the power to change its mind.

Two Oregon towns make different public choices about cameras while standing on opposite banks of the same river.

Democratic custody includes the authority to continue a surveillance program and the authority to end one.

Oregon wrote a floor into law

Governor Tina Kotek signed Senate Bill 1516 on March 31, 2026. The law took effect immediately and established Oregon’s first statewide statutory structure for law-enforcement use of automated license plate readers.

The law does not bless every use. It begins with prohibition, then identifies bounded exceptions.

Agencies may use plate readers for specified purposes, including vehicles connected to crimes, qualifying warrants, missing or endangered people, uninsured or unregistered vehicles, parking operations, and secured areas. The law limits ordinary retention to thirty days unless the data relates to an ongoing criminal investigation or court proceeding.

The law also requires every search to leave a trail. The search log must identify the authorized user, employing agency, search inputs, date and time, case number where applicable, law-enforcement purpose, and offense under investigation. Before an officer stops a vehicle based on an alert, the officer must visually confirm the plate, issuing state, and other vehicle characteristics, then ensure that the stop complies with policy and law.

That sequence matters.

The machine identifies a possible match.

A human verifies the physical vehicle.

An officer determines whether the law permits action.

Later review examines whether the officer used that authority properly.

Oregon also requires vendors to provide monthly and quarterly audits. Those records include camera locations, search totals, authorized users, unique vehicles captured, alerts, outside-agency searches, stated purposes, and the number of cameras accessed. Agencies must make audit results publicly available within two days, subject to required privacy protections.

Before deployment, agencies must publish policies covering security, hot-list quality, training, retention, destruction, audits, and authorized use. New or renewed contracts must establish agency ownership of the data, route legal demands through the agency, require encryption and criminal-justice security controls, and preserve vendor liability for misuse or improper release.

The law limits unrestricted access by non-Oregon government entities. An Oregon agency may conduct a bounded search for an outside entity when the search serves an authorized law-enforcement purpose, but the agency must log the requester, purpose, and scope.

These provisions deserve defense.

They transform a vendor product into a governed public instrument. They place purpose, retention, access, logging, human confirmation, audits, contracts, and remedies inside law.

But a floor is not a conscience.

A statute can command a system to stop. The system must still stop.

Eugene learned that distinction when police confirmed that a Flock camera became active after the city had ordered the cameras shut down. The police chief attributed the event to a reboot following an error message and said he found no indication that police used the resulting data. He nevertheless concluded that the vulnerability contradicted the council’s direction and helped justify ending the contract.

That episode did not prove conspiracy.

It proved something more ordinary and more useful: democratic commands require technical enforcement.

A council vote does not unplug a camera, nor a policy delete a record. Contract do not detect violations by themselves. A right without an executable control remains a request addressed to power.

Citizens discover that a surveillance cable remains active even after they pull a lawful public shutdown lever.

Democratic rules matter only when the technology carries them into operation and leaves evidence when it does not.

Let AI carry the bureaucracy, not the badge

Ethical AI can make surveillance safer because surveillance produces more administration than humans can consistently perform.

Every search needs a purpose. Every purpose needs authority. Every hot-list entry needs current information. Every retained record needs an expiration. Every sharing relationship needs a boundary. Every alert needs confirmation. Every unusual access pattern needs review. Every deletion promise needs evidence.

People forget.

Institutions drift.

Software can remember the rule.

An ethical control system could require a structured case number, offense category, investigative purpose, requesting agency, authorized user, target scope, and legal basis before executing a search. It could reject expired cases, missing purposes, forbidden categories, stale hot-list entries, or requests that conflict with Oregon law.

It could also watch the watchers.

The system could flag repeated searches for one plate across unrelated cases. It could identify unusually broad geographic queries, off-hours access that conflicts with assignment, repeated use of vague purposes, sudden expansion of data sharing, searches involving protected locations, or attempts to reconstruct a person’s movements through a series of individually narrow requests.

But the AI must speak with discipline.

It may say, “This search pattern departs from policy and requires independent review.”

But we must decide, “This officer is corrupt.”

It may say, “This vehicle may match a current stolen-vehicle record.”

It is up to us to conclude, “This driver is a thief.”

It may say, “The evidence supports preserving this image under a specific case hold.”

We must be the one’s collectively to decide how long we retain the data.

NIST’s AI Risk Management Framework emphasizes defined roles, intended-use boundaries, testing, monitoring, incident response, appeal, override, stakeholder feedback, and recourse. It also warns that deployed systems can behave unexpectedly and that organizations must assign people to monitor and answer for those outcomes.

The Justice Department’s review of AI in criminal justice similarly treats automated license plate recognition as an increasingly common surveillance tool whose use raises questions about accuracy, privacy, civil rights, and civil liberties.

Those principles point toward a simple division of labor.

Let AI perform the tireless work:

  • verify required fields;

  • compare requests with policy;

  • detect inconsistencies;

  • monitor access;

  • enforce time limits;

  • prepare audits;

  • redact public reports;

  • preserve tamper-evident receipts;

  • route anomalies to independent review.

Let humans retain the sovereign work:

  • determine whether the evidence justifies further action;

  • approve exceptional access;

  • decide whether to stop, search, arrest, charge, disclose, or prosecute;

  • hear challenges;

  • repair errors;

  • answer to the public.

The AI should carry the bureaucracy. It should not carry the badge until collective humanity governs such an eventuality.

A transparent AI clerk checks surveillance records while a bright boundary separates it from badges, gavels, and arrests.

Ethical AI can enforce procedure and expose misuse. It must not decide guilt or exercise coercive authority.

A free city knows how to forget

A democracy should not remember every innocent journey merely because storage has become cheap.

Oregon permits agencies to retain ordinary plate data for as long as thirty days. That limit creates a ceiling, not a command to use every day available. Communities can choose less.

Flock now says it recommends and defaults new configurations to seven-day retention, with a separate Evidence Mode for preserving particular records connected to active investigations. The company has also announced required case codes, offense filtering, audit assistance, proactive lockouts, multi-factor authentication, and additional security review. These remain vendor representations that communities should test through contracts, audits, and independent evidence rather than accept on faith.

Seven days may still prove too long for some communities and too short for some defined investigations. The right number should follow purpose, evidence, and local consent.

The governing principle matters more than the vendor default:

Most observations should die.

A camera read that does not match a lawful alert, support a documented investigation, or enter a narrowly defined evidence hold should disappear automatically. The system should preserve proof that deletion occurred without preserving the underlying journey.

A longer hold should require a specific case, a named official, a defined evidentiary purpose, a new expiration date, a reason code, and review of privacy consequences.

The archive should not grow simply because nobody remembered to prune it.

Forgetting is not a defect in public safety.

Forgetting is one of liberty’s oldest defenses.

A state that remembers every movement eventually changes the meaning of movement itself. The citizen begins to travel beneath a second road, invisible but permanent, made from records that outlive the reason anyone collected them.

Public memory should serve an investigation.

It should not become the atmosphere.

Ordinary vehicle traces dissolve into rain while a small evidence-bound record remains inside a sealed timed lantern.

A public system should preserve evidence deliberately and release ordinary life from its memory.

A match is not a biography

Not every search carries the same moral weight.

A real-time alert that a currently stolen vehicle passed one camera does not equal a retrospective reconstruction of someone’s movement across a month.

The first event may help an officer act quickly.

The second can reveal a life.

Oregon should govern those actions at different levels.

A current alert tied to a stolen vehicle, active qualifying warrant, or missing person can trigger immediate human review and the visual confirmation the law already requires. A historical search across several cameras should require supervisory approval and a specific investigative threshold. A pattern-of-life search that attempts to reconstruct where a person traveled, whom they visited, or which sensitive places they entered should require a warrant or comparable court order, except during a narrowly defined emergency involving imminent danger.

The technical system should enforce those distinctions.

It should not allow an investigator to transform a single-camera inquiry into a movement map by widening the search one harmless-looking step at a time.

Ethical AI can detect that sequence because it sees the whole request history. It can recognize when ten narrow searches have become one broad surveillance act.

It should then stop the query and ask for the authority that the combined effect requires.

The law often evaluates actions one transaction at a time.

Power accumulates across transactions.

The control system must see both.

Protect the places where freedom gathers

Some places carry more than geographic meaning.

  • A school holds children.

  • A shelter holds people escaping danger.

  • A clinic holds decisions that belong to patients.

  • A house of worship holds conscience.

  • A union hall holds collective action.

  • A newsroom holds confidential sources.

  • A political gathering holds dissent.

  • An addiction-treatment center holds recovery.

  • An immigration law office holds people whose safety may depend on the line between local service and federal enforcement.

The presence of a vehicle near one of these places does not establish wrongdoing. It may not even establish who drove it.

Oregon should identify sensitive-location classes and require stronger controls when a search touches them. The location itself must never become a reason to infer criminality, extremism, immigration status, health status, religion, political belief, or vulnerability.

This does not grant immunity to violence.

Police must remain able to investigate serious crime near a school, clinic, shelter, house of worship, or protest. The government should simply carry a heavier evidentiary burden before it turns the geography of lawful association into an investigative map.

A sensitive-location search should require a specific case, a narrow purpose, independent approval, shorter retention, and later review.

The system should protect lawful association without creating sanctuary for harm.

A mature democracy can hold both duties at once.

Surveillance lines stop at a civic boundary before schools, clinics, shelters, worship, labor, journalism, and peaceful assembly.

A lawful destination should not become a character judgment. Sensitive places require stronger evidence, narrower access, and independent review.

“Bad actor” must describe conduct, not essence

The phrase bad actor tempts the speaker to skip the evidence.

It takes a bounded event and turns it into a permanent kind of person.

An ethical surveillance system should never ask AI to decide who looks dangerous, which neighborhood produces criminals, which associations imply guilt, or which beliefs deserve suspicion.

It should identify specific evidentiary conditions:

  • A vehicle appears on a current stolen-vehicle list.

  • A plate matches a qualifying warrant record.

  • A specifically described vehicle may relate to a reported crime.

  • A vehicle may carry a missing or endangered person.

  • A current investigation, supported by lawful process, requires a narrow historical search.

  • An authorized user may have searched outside policy.

Those conditions can justify attention.

They cannot establish moral identity.

A stolen-vehicle alert may involve an innocent owner, a cloned plate, a data-entry error, a family member, an outdated report, or a sensor mistake. A camera read can establish that a device captured an apparent match at a time and place. It cannot establish who drove, what that person knew, what they intended, or whether a court will find them responsible.

The system must carry that uncertainty forward.

It should never allow a probability to harden into a noun.

Candidate is not culprit. Alert is not arrest. Association is not automatic guilt and receipt is not truth.

Flock is a vendor, never a constitution

Flock has announced reforms that deserve serious examination. The company now describes seven-day default retention, evidence-specific preservation, mandatory case codes, offense filters, audit assistance, proactive account lockouts, multi-factor authentication, independent security review, customer ownership, and customer control over sharing.

A democracy should welcome a vendor that improves its controls.

It should not confuse improvement with self-government.

The ACLU has continued to criticize Flock’s representations, data-sharing structure, product expansion, and credibility. Its objections show why vendor claims must remain contestable and why elected governments must rely on law, contract, inspection, and remedies rather than promotional assurances.

The proper response does not require blind trust or automatic rejection.

Write the promise into the contract.

Define the promise in machine-readable terms.

Test it.

Audit it.

Publish the result.

Create liability when the promise fails.

Allow the community to terminate the relationship.

A company may build the camera.

It may not write the constitution around the camera.

A camera vendor stands beneath a public charter while contracts connect the device to audits, deletion, liability, and an off switch.

Vendor safeguards matter only when public law, enforceable contracts, independent evidence, and democratic withdrawal make them real.

The people must hold the off switch

A city should never purchase surveillance as though it were buying office furniture.

Technology changes. Vendors change. political conditions change. Search capabilities expand. Sharing relationships widen. New agencies ask for access. Artificial intelligence discovers new ways to combine old records.

Consent must therefore expire.

Every public surveillance program should end automatically unless the governing body renews it after a public review.

Renewal should require evidence of investigative value, local accuracy, false-alert rates, complaints, misuse findings, data-sharing relationships, retention practices, civil-rights effects, less intrusive alternatives, and the full cost of operation.

Police should testify.

Crime victims should testify.

Public defenders should testify.

Civil-liberties advocates should testify.

Technologists should testify.

Immigrant communities, health-care providers, journalists, labor organizations, neighborhood groups, and ordinary residents should testify.

Then elected officials should vote again.

Florence exercised that power when it renewed the cameras.

Albany exercised that power when it ended them.

Eugene exercised that power when it concluded that the technical and trust failures had exceeded what the city would accept.

The off switch does not represent hostility to public safety.

It represents the final proof that the system belongs to the public.

A government that cannot turn a surveillance tool off does not govern the tool.

The tool governs the government.

Oregon residents and public officials share control of a large switch governing a network of roadside cameras.

Democratic custody becomes real when the public can continue a system, narrow it, or bring it to an end.

The public eye must kneel

At night, the roadside camera continues its vigil.

It sees a plate cross the rain.

That observation may help recover a stolen car. It may help locate a missing person. It may give investigators the lead that lets them act before a trail disappears.

That is real public value.

The same observation may also enter a network that remembers too much, shares too broadly, interprets too confidently, or survives long after its purpose has died.

That is real public danger.

We should not pretend that safety and liberty occupy opposite shores. A person whose child has disappeared wants investigators to move quickly. A person who seeks medical care, attends a protest, enters a shelter, meets a confidential source, or visits a house of worship wants the government to leave that lawful movement alone.

Both people belong to the public.

A worthy system must answer to both.

Ethical AI can help. It can carry the dull weight of governance that institutions too often neglect. It can require a valid purpose, detect a contradiction, expose a suspicious search, enforce a deletion date, compare an alert with current records, prepare an audit, and preserve the evidence of misuse.

Then it must yield.

A person must decide whether evidence justifies action.

An independent reviewer must examine disputed use.

A court must determine whether government crossed a legal threshold.

The public must decide whether the system may continue.

Surveillance becomes legitimate only when it kneels:

  • to law;

  • to particular purpose;

  • to limited memory;

  • to independent audit;

  • to human verification;

  • to due process;

  • to correction;

  • to public explanation;

  • to the right of challenge;

  • to the people’s power to end it.

Not privacy without safety.

Not safety without liberty.

A public eye whose eyelid belongs to the people.


Corrections and right of reply

UVLM invites documented corrections, relevant public records, technical documentation, agency responses, vendor responses, and good-faith disagreement from people and institutions discussed in this article.

A correction request should identify the disputed passage, explain the requested change, and provide supporting evidence. Factual correction does not require agreement with the article’s policy position.


Claim-boundary note

This article offers public-policy commentary, not legal advice.

It does not claim that every Flock deployment operates identically, that every plate-reader search violates the law, that every agency has misused the technology, or that automated license plate readers independently reduce overall crime.

It distinguishes official law, municipal decisions, police operational reports, vendor representations, civil-liberties objections, documented technical failures, and the author’s governance proposals.

The article uses bad actor only as shorthand for conduct relevant to a specific, lawful, evidence-bound inquiry. It does not authorize AI to classify a person’s character, guilt, credibility, political legitimacy, dangerousness, mental state, or moral worth.


Works Consulted

Research cutoff: August 17, 2026.

The article treats enacted law as legal authority within its stated scope; municipal and police statistics as agency-reported operational evidence; Flock materials as vendor representations requiring verification; ACLU materials as civil-liberties advocacy and legal analysis; and journalism as secondary reporting. No single source independently validates the article’s complete governance model.

Oregon law and legislative record

  1. Oregon Legislative Assembly. Senate Bill 1516, 2026 Regular Session, enrolled as Chapter 77. The official measure restricts authorized ALPR uses, limits ordinary retention to thirty days, regulates sharing, requires agency policies, governs vendor access, and creates civil remedies.

  2. Oregon Capital Chronicle, republished by OPB. “A New Oregon Law Regulates Police Use of License Plate Readers. Here’s How It Works.” The report covers Governor Tina Kotek’s March 31 signature, immediate effect, audit requirements, data-sharing controls, and competing assessments of the final law.

  3. ACLU of Oregon. Testimony in Opposition to Senate Bill 1516-A. The testimony describes privacy, association, immigration, reproductive-care, and misuse concerns and cites reported cases involving abortion-related searches and personal misuse by an officer.

  4. Oregon Association Chiefs of Police and Oregon State Sheriffs’ Association. “Responsible Use of License Plate Reader Technology: Privacy, Security, and Public Safety.” This testimony supplies the law-enforcement case for bounded ALPR use and examples offered in support of the technology.

Florence

  1. City of Florence. January 5, 2026 City Council agenda packet and police memorandum. Florence reported alerts involving 47 stolen plates, 36 stolen vehicles, and 20 missing people, along with vehicle recoveries, missing-person contacts, suspect identification, and assistance in tracing a homicide suspect’s vehicle. The figures remain police-reported operational results rather than an independent crime-reduction study.

  2. City of Florence. “Florence City Council Announces Decision on Flock Cameras.” The city documented the February 2, 2026 public decision and the 3-2 vote to continue using the system.

  3. KLCC. “Florence Votes to Keep Automatic License Plate Cameras.” The report covers the divided vote, resident opposition, privacy concerns, and the city’s choice to continue its contract.

Albany, Eugene, Springfield, and Bend

  1. KLCC. “Albany City Council Decides Against Reactivating Flock Camera.” Albany voted 4-2 to end its use of the system after public meetings and debate over safety, privacy, federal access, and surveillance of law-abiding residents.

  2. KLCC. “Eugene Police Confirms Flock Camera Was Turned On After City Ordered the Cameras to Be Shut Down.” Eugene’s police chief said one camera apparently rebooted after an error, reported no indication that police used the resulting data, and concluded that the vulnerability contradicted council direction.

  3. OPB and KLCC. “Eugene and Springfield Both Announce End of Flock Camera Usage.” The two cities announced that they would discontinue the technology after public controversy and governance concerns.

  4. OPB. “Bend Is the Latest Oregon City to Turn Off Flock Cameras.” Bend decided not to renew its pilot contract amid public pressure and concerns about privacy, data collection, and federal use.

Civil-liberties analysis

  1. ACLU of Oregon. “Know Your Tech: Flock.” This public educational resource explains the organization’s concerns about network scale, searchable movement history, images beyond license plates, data sharing, and mission expansion.

  2. ACLU. “Flock Safety Credibility Lost as It Repeatedly Lies to City Councils, Police Departments, and the Public Across the Country.” This advocacy analysis challenges the reliability of Flock’s public representations and argues that governments should not treat vendor assurances as sufficient governance.

  3. ACLU of Oregon. May v. City of Eugene. The case record concerns a public-records dispute over Eugene’s Flock system. It establishes the existence and posture of the litigation, not the truth of every broader surveillance claim.

Vendor representations

  1. Flock Safety. “Flock Updates Privacy, Accountability, Security, and Transparency Safeguards.” Flock announced a recommended seven-day default retention period, Evidence Mode, offense filters, required case codes, misuse detection, proactive lockouts, mandatory multifactor authentication, and additional security review. These remain vendor representations whose practical effects require contract and audit verification.

  2. Flock Safety. “Compliance Tools.” The company describes agency-controlled sharing, transparency portals, encryption, case-code and offense controls, audit assistance, and seven-day default deletion.

AI governance and criminal justice

  1. National Institute of Standards and Technology. AI Risk Management Framework Playbook, Govern. NIST recommends defined purposes, risk mapping, monitoring, incident response, stakeholder feedback, recourse, opt-out mechanisms, clear human roles, independent testing, and deliberate decommissioning.

  2. National Institute of Standards and Technology. AI Risk Management Framework Playbook, Measure. NIST recommends measuring human oversight, downstream actions, overrides, errors, complaints, response times, adjudication, and accountable go or no-go decisions.

  3. United States Department of Justice. “Artificial Intelligence and Criminal Justice, Final Report.” The report addresses ALPR as an AI-enabled surveillance practice and frames responsible criminal-justice AI around efficiency and accuracy alongside privacy, civil-rights, and civil-liberties protections.

Previous
Previous

Care Is Infrastructure, Not Sentiment

Next
Next

The Door That Opens Ten More: Tarot, synthetic prophecy, and the cult without walls