InfrastructureIssue 0112 records
Transportation
Transportation AI this cycle is dominated by camera enforcement and staged governance. MBTA began its bus-lane camera education period and Chicago made Smart Streets citywide, while Singapore's false-positive counts and Victoria's audit show what evidence those programs still owe. Signal and intersection AI remains advisory or in proof of concept (Boston, Caltrans, USDOT ISS). Vendor announcements from Ontra, MIA and Masabi lead with unvalidated projections, and Utah's executive order and California's SB 1246 write human accountability into policy.
- Evidence records
- 12
- Cross-source patterns
- 4
- Evidence classes
- 4 government evaluation4 independent reporting2 vendor claim1 government audit1 standards or public-body guidance
- Outcomes
- 7 emerging2 mixed2 cautionary1 effective
- Source freshness
- 6 recent4 new this fortnight1 undated1 older, newly relevant
- Research completed
- 2026-10-30
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Camera enforcement scales faster than published accuracy evidence
MBTA began its education period and Chicago approved citywide expansion this month, both relying on human review, while Singapore's trial showed 257 false positives alongside about 90% detection and Victoria's audit found an AI camera program could not show road-safety benefit and breached retention rules.
Operating questionBefore fines start or zones expand, what recall, false-positive, retention and benefit evidence will the agency publish?
Supporting evidenceMassachusetts Bay Transportation Authority (MBTA)City of Chicago (Mayor's Office, CDOT, Department of Finance) with CTA and NITAMinistry of Transport, Singapore / Land Transport AuthorityVictorian Auditor-General's Office (VAGO)
Signal and intersection AI is still advisory or in proof of concept
Boston's advisory signal recommendations produced modest average gains with about a third reversed; Caltrans adaptive AI signal control is a proof of concept; USDOT's intersection safety program is only beginning prototype testing of automated interventions.
Operating questionWhat validation and fail-safe evidence is needed before AI can change signal operations without an engineer approving each change?
Supporting evidenceEno Center for Transportation (author: former Boston Chief of Streets)California Department of Transportation (Caltrans), Division of Engineering ServicesU.S. DOT Intersection Safety Systems program; Derq with TxDOT, City of Houston, VDOT, Alexandria and Falls Church
Transport AI announcements lead with projections, not measured results
Ontra's ridership and revenue figures are explicitly unvalidated estimates, MIA's savings assume full adoption, and Dayton's test measures accuracy but not productivity.
Operating questionWhich baseline and post-deployment metric will the agency own so a projection can be checked?
Supporting evidenceOntra Mobility with MTA Long Island Rail Road and MTA New York City Transit (Transit Tech Lab)Miami-Dade Aviation Department (Miami International Airport) with KaléGreater Dayton Regional Transit Authority (RTA) with Masabi
Human accountability is being written into policy
Utah's executive order requires human review for rights-affecting decisions, Caltrans layers a directive and GenAI subcommittee over its AI portfolio, and California's SB 1246 makes AV operators accountable to first responders and local jurisdictions.
Operating questionWhich transportation AI uses are rights-affecting or safety-critical, and who is the named human accountable for each?
Supporting evidenceUtah Department of Transportation (UDOT) / Office of Governor Spencer CoxCalifornia Department of Transportation (Caltrans), Division of Engineering ServicesState of California (SB 1246; California DMV to set guidelines)
Evidence ledger
Full record · every source keeps its link and limitations
Artificial Intelligence at Caltrans (Bridging the Gap, October 2026)
Caltrans reports a portfolio-stage view of its AI work: a 2026-2028 Data and AI Strategic Plan, Deputy Directive 54 and a GenAI Governance Subcommittee; adaptive AI signal control (District 5) and AI wrong-way detection (District 3) in proof of concept; and Microsoft 365 Copilot, Traffic Mobility Insights and Road Safety Insights approved for use. It says Road Safety Insights integrates more than 30 datasets and that Traffic Mobility Insights unified seven statewide systems so corridor analysis takes minutes instead of hours.
Why it matters, evidence and limitations
- Why it matters
- A large state DOT publicly staging AI by maturity (discovery, proof of concept, approved) gives peers a template for portfolio governance and shows where operational AI (signals, wrong-way detection) still sits: pre-production.
- Evidence and measured results
- Agency self-report in an internal engineering newsletter. The hours-to-minutes claim has no stated method, sample or baseline; no outcome metrics for signals or wrong-way detection are given.
- Limitations and uncertainty
- Self-reported and promotional in tone (ends with a recruiting pitch); no measured results; publication day not stated beyond October 2026 (date recorded as the first of the month).
MBTA Automated Camera Enforcement Program (bus lanes and bus stops)
MBTA began a public education period on October 1, 2026 for bus-mounted automated enforcement on routes SL4 and SL5, with ticketing from January 1, 2027 and expansion to more routes in early 2027. Cameras capture plates automatically and every image is human-reviewed before a ticket. State law limits data use to bus lane and stop enforcement without a court order and requires face blurring; evidence is deleted 120 days after closure, or kept up to three years if unpaid.
Why it matters, evidence and limitations
- Why it matters
- A concrete statutory and policy template for transit camera enforcement (purpose limitation, face blurring, retention limits, warnings, human review) that RTAs and municipalities authorized under the same 2025 Massachusetts law can reuse.
- Evidence and measured results
- Agency program page describing rules and timeline. No performance data yet because the education period is under way. The page does not name the vendor; trade press (Mass Transit, October 22, 2026) reports it is Hayden AI.
- Limitations and uncertainty
- No accuracy, false-positive or bus-speed results yet; appeals details pending; vendor accuracy unknown.
Chicago makes Smart Streets automated parking enforcement permanent and citywide
City Council unanimously approved Ordinance 2026-0028833 making the Smart Streets camera enforcement pilot permanent and expanding it citywide over two years. The city reports 140,000 warnings and violations since November 2024; 75% were no-fine warnings, 84% of warned drivers did not later receive a fine, 64% of ticketed drivers did not get another, and bike-lane incidents fell from one in four to one in five violations. CTA reports a roughly 26% decline in bus stop violations between May and August 2026, and NITA funded 46 more camera-equipped buses.
Why it matters, evidence and limitations
- Why it matters
- A large U.S. municipal program combining city-vehicle and bus-mounted camera enforcement across bus lanes, bike lanes, meters and street cleaning, now scaling with explicit compliance metrics.
- Evidence and measured results
- City press release with program metrics; no independent evaluation, methodology, or accuracy and false-positive data. Fatality reductions cited are citywide and not attributed to the program.
- Limitations and uncertainty
- Self-reported with advocacy framing; no error rates; behavior-change metrics do not control for other factors.
Singapore MOT reply: AI-detected bus lane offence accuracy, officer review and appeals
In a trial analyzing over 800 hours of bus camera footage over three months, LTA's AI detected 415 of more than 450 bus lane encroachments (about 90%) but also produced 257 false positives. All AI-detected violations will still be manually reviewed before enforcement, contests and appeals stay with human officers, and the minister said LTA aims to reduce the need for human oversight as accuracy improves.
Why it matters, evidence and limitations
- Why it matters
- Rare published recall and false-positive counts for bus-mounted AI enforcement give U.S. transit agencies (MBTA, CTA) a concrete benchmark for what to demand from vendors and why human review remains necessary.
- Evidence and measured results
- Government parliamentary answer with trial counts; denominators are approximate ('more than 450'), with no precision figure and no breakdown by conditions.
- Limitations and uncertainty
- Singapore's legal and enforcement context differs from U.S. states; trial scope limited. False positives were roughly 38% of all detections (257 of 672), which is our calculation, not the ministry's.
Reducing the Harm Caused by Distracted Drivers (Victorian Auditor-General audit of an AI camera program)
VAGO found the department does not know whether the AI distracted-driver and seatbelt camera program reduces crashes, serious injuries or deaths: no baseline data, changed benefit measures and inconsistent reporting. Image verification works as intended (per 1,000 vehicles photographed, AI passed 147 to human review and 6 infringements resulted), but images were kept longer than regulations allow, the vendor archived images for 7 years until February 2026, self-reported vendor KPIs were not verified, and the department could not confirm which AI software version was deployed from March 2024 until December 2025.
Why it matters, evidence and limitations
- Why it matters
- Why now: U.S. agencies are launching or expanding AI camera enforcement this month (MBTA education period from October 1; Chicago citywide expansion October 14). This audit is a detailed public-sector checklist for evaluation baselines, retention, vendor KPI verification and model version control that U.S. programs can apply before scaling.
- Evidence and measured results
- Independent government audit with program data (308 sites, 188,493 infringements, July 2023 to September 2025) and five recommendations (four accepted in full, one in part).
- Limitations and uncertainty
- Australian legal context; the audit concerns evaluation and administration, not detection accuracy itself; published June 2026, before this edition.
Giving Boston the Green Light: lessons from AI signal-timing recommendations
Boston implemented Google Project Green Light recommendations at more than a quarter of its 900 signalized intersections, averaging a 13% delay reduction at modified intersections (27% at several Back Bay intersections), but reversed changes at about a third of locations for lack of improvement or negative impacts. The AI sometimes misread phasing at complex intersections, does no corridor or network analysis, surfaced equipment failures rather than new insights at some sites, and cannot resolve pedestrian-versus-vehicle value tradeoffs.
Why it matters, evidence and limitations
- Why it matters
- Why now: state DOTs such as Caltrans have adaptive AI signal control in proof of concept this month; Boston's account shows what advisory, human-reviewed signal AI does and does not deliver.
- Evidence and measured results
- First-person practitioner account with city-reported averages; no published methodology or independent verification of the 13% figure.
- Limitations and uncertainty
- Author is a former city official; impact measurement is vendor-provided; localized optimizations may not yield network benefit.
How Dayton is using AI to accelerate transit analysis
Greater Dayton RTA is testing Masabi's natural-language analytics tool over ridership, ticketing and fare-validation data. Per Masabi's CEO, a transfer-pattern analysis that had taken significant manual effort took five minutes and matched earlier results. Dayton is also examining route ridership and fare-media use for its network redesign, but testing focuses on accuracy rather than quantifying productivity gains. The work is part of a four-agency incubator that also includes ML-based revenue-protection research with human review of every flag.
Why it matters, evidence and limitations
- Why it matters
- Mid-size transit agencies are early adopters of self-service natural-language analytics on fare data; the accuracy-first test design and rider-data safeguards are a useful model.
- Evidence and measured results
- Trade reporting relying largely on vendor CEO quotes; one anecdotal time comparison; no agency-published evaluation.
- Limitations and uncertainty
- Single anecdote; productivity not measured; vendor-sourced claims.
Ontra Mobility shares Transit Tech Lab proof-of-concept results with MTA LIRR and NYC Transit
In an eight-week Transit Tech Lab proof of concept, Ontra applied its AI planning platform to LIRR weekend timetables (Huntington and Ronkonkoma branches) and a Brooklyn bus network redesign. Ontra estimates a cost-neutral LIRR redesign could save about 177,000 passenger-hours a year and the Brooklyn redesign could add about 2.1 million annual riders and $3.5 million in fare revenue, and states these are its own estimates not validated by the agencies.
Why it matters, evidence and limitations
- Why it matters
- Shows how agencies use structured accelerator pilots (17 companies, MTA and Port Authority) to test AI planning tools, and illustrates the gap between modeled benefits and validated outcomes.
- Evidence and measured results
- Vendor press release with self-reported model estimates explicitly not validated by the agencies.
- Limitations and uncertainty
- All benefits are modeled estimates; no implementation occurred; vendor source.
MIA becomes the world's largest cargo airport to launch an AI-powered digital platform
Miami International Airport launched an AI-powered Airport Cargo Community System from Kalé, announced October 27 at the TIACA Air Cargo Forum, as a shared portal for forwarders, handlers, truckers, carriers and ground handlers to improve information sharing, planning and truck scheduling. MIA handled 3.4 million tons of freight in 2025 with more than 1,600 daily truck trips. Projected benefits (up to $250 million in annual operational savings and 5,900 tons of CO2) assume full adoption and rest on 'modest estimates' of 15% paper-handling savings and two hours of truck turnaround.
Why it matters, evidence and limitations
- Why it matters
- A public airport deploying a shared AI-enabled data platform across private cargo partners, a model for airports and port authorities managing landside truck congestion.
- Evidence and measured results
- County press release; benefit figures are conditional projections based on vendor estimates, not measured results. Classified as vendor-claim because the quantified benefits derive from the vendor's model.
- Limitations and uncertainty
- The release does not specify what the AI does; benefits assume full adoption; no baseline published.
Gov. Cox signs Executive Order 2026-08 on AI use in state government (UDOT employee guidance)
UDOT told employees that Executive Order 2026-08, rooted in Utah's Pro-Human AI Initiative, sets a statewide framework for AI across agencies: AI supports rather than replaces human judgment, decisions affecting individual rights must include human review, statewide AI literacy training and extra training for agency AI leads will be offered, and agency AI leads will share practices under data and privacy safeguards.
Why it matters, evidence and limitations
- Why it matters
- Shows how a state-level AI order cascades into a DOT's workforce guidance, including human review for rights-affecting decisions that would apply to enforcement and permitting uses.
- Evidence and measured results
- Agency summary of an executive order; the order's full text was not inspected for this item.
- Limitations and uncertainty
- Summary of an order, not the order itself; no implementation details or timelines.
California SB 1246: robotaxi operators face fines for blocking first responders
California enacted SB 1246, effective July 2028, requiring AV companies to use only U.S.-based remote drivers with U.S. licenses, notify local jurisdictions of vehicle locations and status during system-wide failures, and provide local incident technicians; companies can be fined if an AV blocks first responders during an emergency for more than 30 minutes. The DMV will set guidelines including response times.
Why it matters, evidence and limitations
- Why it matters
- Gives cities, transit agencies and DOTs a template for operational accountability of autonomous fleets on public roads, including local notification duties during failures.
- Evidence and measured results
- News reporting on an enacted law with legislator and operator statements; bill text not inspected for this item.
- Limitations and uncertainty
- Effective July 2028; key parameters such as response times are not yet defined.
Derq to test AI-powered crash prevention at U.S. intersections (USDOT Intersection Safety Systems)
USDOT selected Derq as one of three teams (with teams led by UCLA and the University of Michigan) in the Intersection Safety Systems program to build and test AI that predicts conflicts such as red-light running and pedestrian crossings and triggers warnings, roadside signs or signal-operation changes. Testing starts at Virginia Smart Roads and TTI's RELLIS Proving Grounds before possible operational intersections. The solicitation estimated $5.5 million for Phase 1 and $11.5 million for Phase 2; no crash-reduction results exist yet.
Why it matters, evidence and limitations
- Why it matters
- Federal program with state DOT and city partners testing predictive, closed-loop intersection safety, including signal changes, which sets future procurement and validation expectations.
- Evidence and measured results
- Trade article hosted on the vendor's site; program facts cite USDOT materials; no performance data.
- Limitations and uncertainty
- Prototype stage; detecting conflicts is not the same as preventing crashes, as the article notes.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Government evaluation
- A public body’s measured evaluation or documented pilot.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Government audit
- An oversight review of performance, controls, or operations.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.