One California town's police department discovered that its AI-powered license plate cameras produced incorrect readings in 71% of cases where the system flagged a vehicle as stolen or linked to a felony. The finding, based on an internal review of 1,427 alerts from Flock Safety cameras in Roseville, casts doubt on the reliability of automated surveillance tools increasingly adopted by law enforcement agencies across the country.

What You Need to Know

Flock Safety's cameras use computer vision to read license plates and cross-reference them with databases of stolen vehicles or wanted suspects. The Roseville review covered a specific period in 2024 and examined every alert the system generated. A 71% false positive rate means officers responded to hundreds of erroneous hits, wasting time and potentially eroding public trust. The company has not publicly disputed the findings.

How the Review Worked

Roseville police manually verified each of the 1,427 alerts the Flock cameras produced over a set timeframe. Officers found that in 1,014 cases, the camera had misread the plate. The errors ranged from partial matches to completely different characters being captured. Only 413 alerts correctly identified a vehicle of interest.

The department conducted the audit as part of its oversight of the surveillance system, which was installed to help recover stolen cars and locate suspects. Police officials have not said whether they plan to adjust how the technology is used based on the results.

Accuracy Concerns With AI Surveillance

The Roseville case is not isolated. Studies of automated license plate readers in other jurisdictions have also found high error rates, particularly in low-light conditions, bad weather or when plates are dirty or obscured. Critics argue that the technology's shortcomings are often downplayed by vendors and that police departments deploy it with insufficient independent testing.

  • False positives waste resources: Officers responding to incorrect alerts spend time that could be used on verified leads or community policing.
  • Privacy implications: Even mistaken readings capture the location and movement of innocent drivers, creating a database that can be searched retroactively.
  • Vendor accountability: Flock Safety markets its cameras as highly accurate, but independent audits like Roseville's suggest otherwise.

Why This Matters

The Roseville audit arrives as police departments nationwide expand their use of automated surveillance. Flock Safety alone operates in more than 4,000 communities. If a 71% error rate is typical, it means thousands of innocent drivers are being flagged daily, and officers are being dispatched on false leads. The consequences extend beyond inefficiency: each false alert carries the risk of a traffic stop that could escalate or disproportionately affect minority communities. Without mandatory independent testing and transparent reporting, cities have no reliable way to assess whether these systems work as advertised. The burden now falls on local oversight bodies to demand hard data before renewing contracts.