What the camera sees
A vehicle, plate, time, date, location, direction, and sometimes make, model, colour, body type, damage, decals, roof racks, or other visible details.
AI, privacy, cameras, cookies, and data centres
Surveillance: what changes and what doesn’t
Licence plate cameras are a legitimate security and police tool when used for a clear purpose. They are not the same thing as the commercial tracking system already living in phones, browsers, apps, cookies, ad networks, location data, and AI-driven advertising.
Where I start
| Question | Practical answer |
|---|---|
| Are licence plate cameras automatically bad? | No. Used for security, stolen vehicles, hit-and-runs, missing persons, and serious crime, they are useful. |
| Do they know who you are? | Not by themselves. They see a plate, vehicle, time, date, location, and direction. Identity requires linkage to other records. |
| Where does AI fit? | AI reads plates, classifies vehicles, searches visible features, matches hotlists, and makes footage easier to search. |
| What is the real risk? | Retention, broad sharing, weak audits, bad database linkage, and people treating AI matches as proof. |
| What is the bigger daily privacy issue? | Phones, browsers, apps, cookies, ad networks, pixels, location data, purchase signals, and AI advertising systems. |
| Do AI data centres create surveillance? | No. They do not invent cameras, cookies, or adtech. But more compute can make search, profiling, video analytics, and prediction easier to scale. |
How I see the risk
The mistake is lumping every camera, cookie, police tool, data broker, and AI model into one monster. They are different risks.
| System | Danger | Upside | Why |
|---|---|---|---|
| ALPR used properly in Canada | 1–2/10 | 8/10 | It helps identify vehicles tied to real security incidents or crimes. With notice, limited access, short retention, and audit logs, the risk is low. |
| ALPR with sloppy implementation | 4–5/10 | 6/10 | The concern is not conspiracy. It is poor compliance: too much retention, too many users, vague sharing, or weak audits. |
| Database creep | 5–7/10 | Varies | Bad plate reads, stale records, borrowed vehicles, old addresses, and weak links can create confident-looking but wrong conclusions. |
| Full surveillance stack | 8–9/10 | Low unless tightly controlled | Plate readers plus face recognition, phone location, payment data, adtech profiles, identity databases, and broad sharing is a different system. That is not a plate camera by itself. |
| Commercial phone/browser/adtech tracking | 7/10 | Mostly convenience and advertising | This is the tracking system most people already live with every day. It is quiet, commercial, and usually much less visible than a camera on a pole. |
Licence-plate cameras
What it does not know on its own
Your name, address, spouse, job, politics, medical history, private habits, or why you were there.
Why people use them
They can help identify stolen vehicles, hit-and-run vehicles, suspect vehicles, missing-person alerts, access-control issues, and repeat theft patterns.
A camera seeing a vehicle is not the same thing as knowing a person. The sensitive step is linking that observation to other data.
Commercial uses I can live with
- Parking lots, gates, warehouses, malls, condos, construction sites, industrial yards, and private property security.
- Access control, theft prevention, trespass response, stolen vehicle alerts, incident investigation, and evidence preservation.
- Handing relevant evidence to police when a crime or lawful investigation is involved.
- Using cameras openly, with notice, reasonable purpose, limited access, safeguards, and retention limits.
Commercial uses I would restrict
- Keeping plate histories forever.
- Selling or reusing movement data for unrelated commercial purposes.
- Using private cameras as a general public-road dragnet.
- Sharing broadly without a clear reason.
- Collecting more than needed for the stated purpose.
- Linking plate data to identity or other profiles without a lawful basis.
Police and government
Police use is the clearest legitimate case. If a vehicle is tied to a real crime or public-safety incident, camera data should help police find it.
| Valid use | Why |
|---|---|
| Stolen vehicles | Fast identification and recovery. |
| Hit-and-run or collision evidence | A plate, vehicle description, time, and location can identify the vehicle involved. |
| Missing persons and Amber Alerts | Vehicle sightings can matter quickly. |
| Break-ins, theft, violent crime, suspect vehicles | Camera data can be evidence or a lead. |
| Case-linked investigation | Police can connect the plate to a person through lawful investigative tools. |
The line is simple: use it for real public-safety and investigation purposes. Do not turn it into casual movement tracking.
The risk is the database, not the camera
A licence plate camera is narrow. The risk starts when simple observations are linked to too many other systems and treated as if the result is perfect.
A plate read may be wrong. A registration may be stale. A vehicle may have been sold. The driver may not be the owner. An address may be old. A vehicle may be borrowed. An AI tool may connect dots that should not be connected.
This is often a garbage-in, garbage-out problem before it is a surveillance-state problem.
Most day-to-day tracking is already commercial
The average person is probably more exposed by their phone and browser than by a licence plate camera.
What gets collected now
Location, searches, clicks, purchases, site visits, app use, videos watched, ads paused on, loyalty cards, device IDs, pixels, and cookies.
Where AI fits
Prediction: what you might click, buy, watch, believe, ignore, or respond to next.
Why that matters
One camera sees one event. Adtech can build a profile across many apps, sites, devices, purchases, and accounts.
The visible camera is not always the biggest privacy issue. The quiet commercial profile often knows more.
Ads, cookies and compute
AI did not invent online tracking. It made the tracking economy faster, more predictive, and more profitable.
| Older version | AI-assisted version |
|---|---|
| A cookie says this browser looked at roofing. | Models predict whether this person is a homeowner, in-market, likely to click, likely to buy, and worth bidding on. |
| A site shows the same ad again. | Ad platforms run real-time auctions, optimize bids, rank creative, predict conversion value, and target similar users. |
| A camera records a plate. | Computer vision reads the plate, classifies the vehicle, searches features, triggers alerts, and makes stored video easier to query. |
| A person reviews footage manually. | Search tools can find visible features such as a white pickup, ladder rack, decal, or damage across large stores of video or plate reads. |
Does a new AI data centre change this?
No, they do not invent surveillance. Cameras, cookies, licence plate readers, ad targeting, smartphones, and police databases already exist.
Yes, they can increase scale. More compute can make it cheaper and easier to store more data, search more video, classify more images, run more prediction models, and connect more systems.
The practical answer: new AI data centres are not automatically surveillance infrastructure, but they can power more surveillance-style workloads if rules are weak.
Rules I would put around it
The answer is not panic or a blank cheque. The answer is purpose, limits, auditability, and consequences.
ALPR rules
Public notice, defined purpose, minimal collection, short retention for non-hit data, restricted access, case-linked searches, audit logs, and no unrelated commercial resale of plate histories.
Database rules
No unrestricted linking just because it is technically possible. Require human review, treat AI matches as leads, and keep error-correction paths.
Adtech rules
Meaningful opt-in consent, simple opt-out, no dark patterns, no sensitive profiling, and no cross-site tracking hidden behind fake consent.
Police rules
Use the tool for public safety and crime. Log access, review audits, tie searches to a real purpose, and punish misuse.
Commercial rules
Use cameras for security, not marketing creep. A business camera should protect the site, not become a side business selling movement histories.
Data centre rules
Privacy rules should follow the workload. If public approvals support AI infrastructure, demand data residency, lawful access controls, auditability, and no backdoor laundering of Canadian data.
Where I land on it
| Claim | Fair answer |
|---|---|
| “Flock cameras are a surveillance state.” | Not by themselves. Used properly, they are a security tool. |
| “Police should not use this.” | Police use for actual crimes and public safety is valid. |
| “The camera knows who I am.” | Not by itself. It needs linkage to other records. |
| “The real privacy issue is AI.” | AI makes data easier to search, classify, predict, and link. That is why rules matter. |
| “The bigger daily threat is commercial tracking.” | Often yes. Phones, browsers, apps, cookies, pixels, platforms, and ad networks already profile people at scale. |
Useful security tools should be allowed. Criminals should be caught. Private companies should not get unlimited movement histories. Governments should not connect every database just because they can. AI output should not be treated as truth.
Sources and notes
- Flock Safety licence plate reader product information for the company’s description of AI-powered plate reads, vehicle make, model, colour, location, and time.
- Flock Safety data privacy information for the company’s stated position on licence plate data, vehicle characteristics, retention, and privacy controls.
- Office of the Privacy Commissioner of Canada — Guidelines for overt video surveillance in the private sector for private-sector notice, purpose, use, access, security, and retention principles.
- Government of Alberta — Collecting personal information under PIPA for reasonable-purpose limits on private-sector collection, use, and disclosure.
- Government of Alberta — Disclosing personal information under PIPA for disclosure to Canadian law enforcement in connection with investigations likely to lead to law-enforcement proceedings.
- Office of the Privacy Commissioner of Canada — Online behavioural advertising policy position for the view that tracking and targeting for behavioural advertising generally involves personal information and requires meaningful consent.
- Google Ads — Smart Bidding for Google’s explanation that auction-time bidding uses Google AI and machine learning to optimize bids using contextual signals.
- Meta/Facebook Help Centre — How Facebook ads use machine learning for Meta’s description of machine learning in ad delivery and estimated action rates.
- IAB — OpenRTB for the programmatic advertising standard around real-time bidding for individual ad impressions.
- This page separates Canadian privacy principles from U.S. political narratives. U.S. examples may be useful context, but they do not automatically define Canadian law or Canadian project risk.