Only 5.9% of traffic enforcement programmes demand AI. None set a standard for it.
AI in Traffic Enforcement: Authority Mandate or Vendor Offer? A five-year audit of 152 validated traffic enforcement procurement programmes across five markets, three continents and 30 countries, published between June 2021 and May 2026.
Not one of the 152 programmes attaches a performance standard to an AI function.
What the data shows
Mandate strength- Five markets: United States, Europe, United Kingdom, Australia / New Zealand, Canada.
- 152 active contracts awarded between June 2021 and May 2026, across 30 countries.
- The AI language searched for: artificial intelligence, AI, machine learning, deep learning, neural network, computer vision, video analytics, and named AI detection products.
- In 6 of the 7 mandatory programmes (85.7%), the entire evidence is the phrase “artificial intelligence enabled” and nothing further.
- The terms “machine learning”, “deep learning” and “neural network” appear in 0 of 152 programmes.
- 0 of 152 programmes attach a performance standard to an AI function: no accuracy threshold, detection rate, false-positive tolerance, validation requirement, revalidation trigger or audit provision.
AI is not absent from traffic enforcement. In the overwhelming majority of published procurement programmes, the authority simply did not specify or demand it.
Where AI language appears
Rate within each groupAustralia and New Zealand hold 6 of the 9 AI-positive programmes from 11.2% of the dataset.
8 of the 9 AI-positive programmes sit in outsourced or managed service contracts.
Deployment modes overlap: a programme deploying more than one mode is counted in each, so the rows do not sum to 152.
- 22 of 152 programmes (14.5%) are both outsourced and mobile, 8 of those 22 (36.4%) contain AI language.
- Across the other 130 programmes, 1 (0.8%) does.
- 5 of the 8 AI-positive programmes in that group also carry a secondary detection function.
AI language clusters in outsourced mobile programmes, most of which also procure secondary detection. Elsewhere it is close to absent.
Five-year rate
2021 to 2026The share of programmes containing AI language, calculated separately for each year, shows no sustained movement across the study period. The rate in 2025 is no higher than in 2021.
1 of 11
2 of 40
2 of 42
3 of 33
1 of 21
0 of 5
* Partial years: 2021 covers June to December, 2026 covers January to May.
Authority mandate or vendor offer?
Study assessmentIf an authority demanded AI, adoption is led from the top, against a published standard. If a supplier offered it, adoption is pushed from the bottom, against no standard at all.
| Signal | If authorities were leading | What the record shows |
|---|---|---|
| Contract model | AI specified in direct procurements the authority designs | 8 of the 9 AI-positive programmes are outsourced. Of the 70 procurements an authority ran directly, just one names AI, and only as a passing mention. |
| Wording | Function-specific requirements drafted per jurisdiction | 6 of 9 repeat the same phrase, “artificial intelligence enabled”, across independent jurisdictions |
| Test attached | Accuracy, validation and audit terms that can be enforced | 0 of 152 attach any performance or audit term |
| Sensing layer | Requirements covering radar and LIDAR processing as well as cameras | 0 of 152 attach AI to radar, LIDAR or sensor fusion |
| Capability vs requirement | Behaviour detection procured with the technology named | 3 programmes procure behaviour detection and name no AI at all |
These observations are consistent with adoption led from the supplier side rather than specified from the authority side.
This is an inference, not a measured finding: the records show what was written, not who proposed it.
Conclusion
Author’s assessmentEvery AI reference in the dataset points at the camera. None points at the radar or LIDAR sensors. Authorities should extend their AI requirements to the sensor, and state the standard it must meet and how it stays compliant under type approval.
The technology is ready, and it is already producing enforcement evidence. The specification has not caught up. Authorities that write AI requirements now will set the standard the market builds to. Those that do not will buy the capability anyway, on terms written by someone else.
Full report, data and citation
Read the full report
The complete findings report, with all tables, figures and the author's assessment, is free to download. No registration or email is required.
The record-level dataset behind every figure on this page (152 active records, 37 rejected records, Dataset v16.1) is available for review to journalists, researchers and practitioners. Email dudi.cohen@curvidia.com with your organisation and intended use.
Cohen, D. (2026). AI in Traffic Enforcement: Authority Mandate or Vendor Offer? An Analysis of 152 Validated Traffic Enforcement Procurement Programmes, June 2021 to May 2026. Curvidia. Study CV-0226, Dataset v16.1.
Methodology
Dataset v16.1152 validated programmes published between June 2021 and May 2026 across the United States, Europe (excluding the UK), the United Kingdom, Australia and New Zealand, and Canada: 30 countries in total.
189 records were examined. 37 were rejected where the enforcement function could not be confirmed, a framework carried no award, or validation found the record unsupported.
Core automated enforcement: speed, red-light and average speed, in any deployment context, including fixed, mobile, school zone and work zone.
AI applied to any detection function is in scope: camera and image analysis, and radar, LIDAR and other sensor processing. Back-office functions are not assessed.
Artificial intelligence, AI, machine learning, deep learning, neural network, computer vision, video analytics, and named AI detection products, applied to any sensing modality.
Number plate recognition (ANPR or ALPR) on its own is not counted. Plate reading predates the technologies measured here.
- Mandatory: AI specified as a compulsory element of the delivered system
- Required: AI named as a requirement without compulsory framing
- Mentioned: AI referenced with no requirement attached
- None: no AI, machine learning or computer vision language
TED (Tenders Electronic Daily), UK Find a Tender, UK Contracts Finder, SAM.gov and US state portals, AusTender and Australian state portals, GETS (New Zealand), CanadaBuys, MERX and BCBID. Supplier releases, regulatory disclosures and government announcements used for confirmation only. 35 distinct source designations.
Every record is confidence-rated against its source: High 114, Medium 28, Low-Medium 9, N/A 1.
All 9 AI-positive programmes are award notices recorded at High confidence, the study's highest source-verification tier.
- The study measures language in procurement records, not deployed technology. A programme with no AI language may still be delivered using AI, so 5.9% is a floor for capability and a measure only of specification.
- Only publicly documented procurements are included. An AI requirement in an unpublished specification would not be captured.
- All classification decisions were made by a single analyst. Type approval and legal metrology frameworks were not assessed.