Flock Safety’s automated license plate reader system is facing renewed scrutiny after two automotive journalists were stopped by police while driving Jaguar Land Rover press vehicles that had been wrongly linked to a stolen plate entry.
The core problem was strikingly simple: an incomplete National Crime Information Center entry for a missing manufacturer plate was reduced to “34 DTM,” while the actual vehicles being driven carried New Jersey plates “34 10 DTM” and “34 08 DTM.” That partial match was enough to trigger police alerts and, in one case, a coordinated stop involving multiple officers.
For investors following public safety technology, the incidents highlight a central tension in AI-enabled surveillance: even a system marketed as highly accurate can create costly real-world failures when scaled across billions of monthly reads and paired with imperfect source data.
Key Facts
- One stop in Plymouth, Minnesota involved a $155,000 Range Rover that police had tracked for days after a Flock Safety alert tied it to a supposedly stolen plate.
- A second stop in Scotts Bluff, Nebraska involved a $105,000 Range Rover Sport and occurred while the driver’s 14-year-old child was in the vehicle.
- The incomplete NCIC entry listed the missing plate only as “34 DTM,” while the actual press vehicles carried “34 10 DTM” and “34 08 DTM.”
- Flock Safety said its system is roughly 99% accurate and processes about 20 billion license plate reads per month.
- Plymouth’s transparency portal shows the city operates 18 cameras that scanned more than 580,000 plates in a recent 30-day period and generated over 14,800 hotlist hits.
Flock Safety ALPR Error
The incidents turned on the interaction between automated recognition, database quality, and police procedure. In Minnesota, officers boxed in a press vehicle in a retail parking lot, approached with hands near their weapons, ordered the occupants out, and conducted pat-downs before confirming with Jaguar Land Rover that the vehicle was legitimate. The vehicle had been flagged because the system matched the visible characters to a hotlist entry that lacked the full plate number.
Flock Safety defended the camera performance on technical grounds, arguing that the system was operating as designed for partial-plate matching. Company executives also acknowledged that NCIC-originated alerts may require a stricter exact-match standard rather than a looser character-presence test. That distinction matters because partial matching can be useful in investigations, but it also raises the likelihood of false positives when source records are incomplete or inconsistent.
The issue affects several groups at once: law enforcement agencies relying on alerts in the field, automakers supplying press fleets with manufacturer plates, drivers who may be pulled over despite no wrongdoing, and technology vendors whose products are increasingly embedded in public safety budgets. It also underscores that the business risk is not limited to hardware accuracy. Data governance, alert design, operator training, and liability exposure can all shape customer trust and future contract growth.
A camera alert is not probable cause, and when partial data meets automated enforcement, the cost of a false positive can escalate quickly.
Why scale changes the risk profile
At 99% accuracy, an AI-based recognition system sounds highly reliable. But at a processing volume of roughly 20 billion reads per month, even a small error rate can translate into a very large number of mismatches or questionable outputs. The critical investor question is not just headline accuracy, but how often errors progress into enforcement actions, complaints, litigation, or contract reviews.
Municipal usage figures show how quickly these systems can generate action signals. In Plymouth alone, 18 cameras logged more than 580,000 plate scans in a 30-day period and produced over 14,800 hotlist hits. That ratio illustrates both the operational value of broad monitoring and the pressure it places on verification workflows before officers initiate stops.
Implications for Investors
For investors in security technology, smart-city infrastructure, and AI surveillance vendors, these incidents are a reminder that adoption growth can bring regulatory and reputational friction. Automated license plate readers have clear demand drivers, including stolen vehicle recovery, investigative support, and municipal safety spending. But each high-profile false stop can sharpen calls for tighter procurement rules, audit trails, retention limits, and exact-match standards for high-consequence alerts.
The near-term watch points are product changes and policy responses. Flock Safety indicated it is working to correct the underlying report and is discussing ways for incomplete NCIC data to be flagged for officers. If vendors move toward more conservative alert logic for federal hotlists, that could reduce false positives but may also alter product performance metrics that agencies value. Investors should watch whether future contracts emphasize explainability, verification controls, and indemnification clauses as much as camera coverage and read volume.
There is also a broader liability and margin question. If wrongful stops lead to more legal claims or force vendors to invest heavily in compliance, training, and oversight tooling, growth may remain intact while profitability becomes more pressured. On the other hand, companies that can demonstrate lower false-positive rates, better data hygiene, and more transparent audit systems may strengthen their competitive position as public buyers become more selective.
The next phase of the story will likely center on whether agencies and vendors tighten standards for partial-plate matching and how federal database quality controls evolve. For investors, the lesson is clear: in surveillance technology, scale can be an advantage only if precision, governance, and field procedures keep pace.