A stop or an arrest that starts from an identity match fails in one of two places. The comparison can be wrong. Or the record that was compared was never a verified record of that person. The second failure is a database failure. It is not fixed by a better algorithm.
Two cases from this month show the second failure. Neither one is an argument about a vendor’s finances.
On 22 September 2026 the Plymouth, Minnesota city council voted 6 to 1 to end its contract for automated license plate readers and to have the cameras removed. The city has 16 of them. The vote was reported the next day by MPR News and by the Sun Sailor.
The incident the council had in front of it was from late June. Joel Feder was stopped in a Kohl’s parking lot by Plymouth officers after the readers alerted on his plate. His New Jersey manufacturer plate read “34 10 DTM”. The alert matched a shorter plate, “34 DTM”. Plymouth police said the stolen-plate entry in the National Crime Information Center had been made in California on 24 June 2026 with incomplete numbers. Officers got alerts on 26 June and 28 June. The 28 June alert is the day of the stop.
The reader and the database did not agree with the plate on the car. Officers still had a car full of police in a parking lot because a stored string said the plate was stolen. Flock’s own statement, given to FOX 9 and to WCCO, was that an alert is a lead and that officers should verify the plate before they act. The council voted to take the cameras out anyway.
Sources:
On 17 September 2026 Biometric Update reported a federal complaint by Angela Lipps of Elizabethton, Tennessee, against the City of Fargo and Detective Lucas Heck. The complaint seeks $10 million. The Guardian reported the suit the next day.
The complaint says Lipps had never been to North Dakota, did not look like the person on the bank video, and was not identified by any bank witness. It says, on information and belief, that the facial-recognition search was run on the photograph printed on a fraudulent identification card the suspect had used, not on a still from the surveillance video of the person who was in the bank. ABC News reported that Fargo’s police chief later said detectives had wrongly assumed the surveillance photos were part of that search, and that they were not.
That is a source-data failure. A match against the picture on a fake document identifies the document. It does not identify a person who was booked, printed, and charged. Treating that match as probable cause for an arrest warrant is the step the complaint challenges. The suit is a set of allegations. It is not a judgment.
Lipps was arrested in Tennessee in July 2025 and held for about six months before the charges were dismissed, according to the Guardian’s account of the filing.
Sources:
UMbRA is the law-enforcement-sourced identity database behind Safience. A record in it is there because an agency booked, convicted, registered, or listed the person, not because a camera misread a plate and not because a picture was taken from a document the suspect printed. The current size of that file is on the UMbRA page and is updated by the site. This story does not state a number. https://safience.io/umbra/
A match against that file is still a lead. A person checks it before an alert goes to a customer. The Plymouth stop and the Fargo arrest both skipped that kind of check: the stored string, or the stored photo, was treated as the person.
The councils and the courts are not arguing about model accuracy in these two cases. They are arguing about whether anyone can show where the matched record came from, and whether a person looked at it before someone was stopped or jailed.
A face search can do its job perfectly and still put the wrong person in...
Read articleIn the space of three days this month, two local governments walked away from the...
Read articleBy a Biometrica staffer Drug related offenses accounted for almost one-third of all cases reported...
Read article
