Eighteen Years of Nightly Lineups

Every night for the past eighteen years, the Iowa Department of Transportation has scanned the face of every new driver’s license applicant against its entire photo database. When its algorithm decides two faces look alike, an investigator pulls credit histories, utility bills, social media posts, and criminal records into a file. No warrant. No probable cause. No indication of fraud. Just a machine’s say-so.

In response to an open records request, the DOT initially said it didn’t know how many of these investigations led to convictions — or even to license denials. It could confirm only that between January 2022 and November 2025, there were 192 such cases.

The DOT promised to compile outcome data by early January and delivered—albeit a bit late. The spreadsheet tells a remarkable story: of the 192 investigations, 28 led to no action of any kind, and only 14 resulted in criminal charges. The spreadsheet does not record whether any of those charges led to convictions.^[1]

This has been happening since at least 2008. And, of course, the FBI has been tapping into Iowa’s database since at least 2014.

Iowa is not unique, but its story illustrates, in granular detail, how the federal government and state DMVs have quietly built one of the largest surveillance infrastructures in American history — and how the agencies operating it have no idea whether it works.

How It Started

In 2005, Congress passed the REAL ID Act. A year later, the Iowa DOT awarded a $1.4 million contract to Digimarc for facial recognition technology. The system was sold to the public as a fraud-prevention tool, but the contract embedded capabilities well beyond that purpose:

Iowa DOT will implement both “one-to-one” and “one-to-many” facial recognition as part of its driver license enrollment process.

“One-to-one” matching — comparing a renewal photo to the applicant’s prior photo — is the anti-fraud use case the DOT advertised. “One-to-many” matching is something else entirely: it scans each new portrait against the full database of driver’s license images. As the press release noted:

Each night, the Biometric Identification system checks each newly captured portrait against the full database of driver license images as another means to catch attempts by a single individual to get a driver license under multiple names.

The DOT pitched this to the public as a civil anti-fraud measure. But it also built out infrastructure with latent capabilities for criminal investigation— and, as it turns out, civil immigration enforcement—capabilities it would formalize with law enforcement partners in the years to come.

By 2007, the nightly scans were operational. Digimarc, the original vendor, was subsequently acquired through a chain of corporate mergers: first by Safran’s Morpho division (later MorphoTrust USA), then by IDEMIA, which continues to partner with the Iowa DOT as of 2024.

The algorithm has changed hands repeatedly, but neither the DOT nor any external body has ever tested its accuracy for the purpose Iowa uses it — flagging potential fraud from a pool of millions.

The FBI Moves In

In 2014, the Iowa Department of Public Safety signed an agreement with the DOT to pay for an upgrade to its facial recognition system in exchange for access to it.

According to Georgetown Law’s The Perpetual Lineup, this gave DPS authorized personnel the ability to run face recognition searches against Iowa’s driver’s license photos.^[2]

That same year, the DOT signed its first Memorandum of Understanding with the FBI, granting the Bureau access to those same photos through its Facial Analysis, Comparison, and Evaluation (FACE) Services unit. A 2016 GAO report confirmed that the FBI could request searches of Iowa’s driver’s license database.

The Iowa MoU was updated in 2018, superseding the original agreement. Under its terms, the FBI submits a probe photo to the DOT, the DOT searches its facial recognition database, and returns a set of candidate photos and identities. The DOT agreed to process up to 15 photo requests per day, but the number it actually processes is not known.

Iowa is hardly alone. In response to a FOIA request about Iowa’s program, the FBI provided a stack of MoUs from other states instead of the Iowa-specific records that would have been responsive.^[3] Those documents show that at least fifteen states have signed similar agreements with the FBI.

By May 2019, 21 states had partnered with FBI FACE Services, giving the Bureau access to over 641 million photos across all searchable repositories.

The Nightly Dragnet, to Start With

The nightly scan is called the Automated Biometric Identification System (ABIS). It has been running every night for eighteen years, telling the DOT’s Bureau of Investigation & Identity Protection (BIIP) who to investigate next — not because there is any indication of fraud, but purely on the output of a commercial facial recognition algorithm.

These investigations are warrantless and deeply invasive. DOT procedure specifies:

When a driver’s license case is first assigned for investigation, the Investigator shall gather all pertinent documents and information related to the investigation.

This includes the complete driving and vehicle ARTS records, specific driving violation records, Accurint check from DPS, or a CLEAR report from a Bureau investigator and criminal history to start with.

— Iowa DOT, Procedures Related to Driver License Investigations, p. 176

A commercial algorithm — for which we have no data regarding accuracy or bias — says your photo looks like one of millions of others. On that basis alone, a DOT investigator pulls your credit history, utility bills, social media posts, criminal history, and every other detail they can find about you into a file. “To start with.”

Although we still don’t have no data on the DOT’s system’s accuracy, we at least have part of the story now: DOT records show that of 192 investigations opened between January 2022 and November 2025 (excluding 19 still-open cases and one duplicate):^[4]

• 28 investigations (16%) resulted in no action whatsoever. No administrative sanction, no criminal referral. The algorithm flagged these people, an investigator pulled their credit histories and criminal records, and then closed the file. Whether the subjects were ever told they’d been investigated is unknown.

• 130 investigations (76%) resulted in administrative action only — mostly license cancellations, but also 20 cases where the sole action was “Merge Records,” which appears to be database housekeeping.

• 14 investigations (8%) resulted in criminal charges. That is 3.5 per year. The charges are overwhelmingly misdemeanors: “Fraudulent Application for DL/ID,” “False Application for DL/ID,” perjury. A handful involved more serious offenses — identity theft, forgery, fraudulent practice. One case resulted in no Iowa charges at all; the subject was extradited to Nebraska on an existing warrant.

• Zero convictions are recorded. The DOT’s spreadsheet tracks charges filed, not final outcomes. After eighteen years and 192 documented investigations, the agency still cannot say whether a single criminal referral from its facial recognition program has ever resulted in a conviction.

One additional detail stands out: in April 2025, an administrative action reads “Notified Homeland Security and IA DOR.”

The nightly scan, sold to the public as a fraud-prevention tool, appears to have been used for federal immigration enforcement, as those agencies continue to push facial recognition mobile app on local police.

Blind Faith in a Blind System

Facial recognition algorithms are notoriously biased, performing worse on people with brown or black skin than on those with white skin. This is a technical reality, not an advocacy position.

In 2016, the Government Accountability Office examined the FBI’s facial recognition program and made six recommendations. The GAO found that the FBI had tested its system’s detection rate only for candidate lists of 50, and had no data on accuracy for smaller list sizes that users regularly requested. The GAO also recommended the FBI determine whether the external state systems it relies on — systems like Iowa’s — were sufficiently accurate.

The FBI disagreed. It told the GAO that its testing satisfied requirements for providing investigative leads and that the Bureau lacked authority to set accuracy requirements for external systems.

Three of the six GAO recommendations dealt with accuracy—a massive red flag for a system that flags and investigates people based on algorithmic photo matches selected from millions of candidates.

Two more dealt with transparency: the FBI had failed to publish required Privacy Impact Assessments^[5] and a System of Records Notice^[6] — legally mandated disclosures for a program handling tens of millions of Americans’ photos. DOJ eventually published the missing documents, though years late.

By 2019, only one of the six recommendations had been fully implemented. The FBI had begun conducting user audits but still had not tested accuracy for smaller candidate lists, had not assessed external partner systems, and had not conducted the recommended annual operational reviews. The GAO maintained all five remaining recommendations were valid.

Iowa, for its part, did no better. In response to public records requests:

• The Iowa DOT and DPS each confirmed that the FBI never shared results of any accuracy tests.
• The Iowa DOT, despite having a “facial recognition analyst” on staff, never performed its own accuracy testing.
• Neither the DOT nor DPS ever solicited or received reports about the system’s accuracy from the vendor. Their only information comes from general materials provided, or published online, by the vendor.
• DPS “[had] not yet adopted a final policy” governing its use of facial recognition, on the grounds that it was waiting to determine “what uses may be accurate or inaccurate, reliable or unreliable, appropriate or inappropriate.”^[7] That was in 2016. A decade later, it still hasn’t looked at system accuracy—knowledge it claims is a prerequisite to regulation.

The FBI did not respond to a FOIA request (other than with the stack of MoUs from other states that were not part of the request).

As the ACLU of Minnesota summarizes: “Facial recognition automates discrimination.” The agencies either believe their algorithm is infallible, or they don’t care whether it’s accurate as long as it gets them an occasional result. Substantial academic and technical literature on algorithmic bias goes ignored, as does the federal government’s own accountability office.

The Bigger Picture: National Mass Surveillance

Iowa’s nightly scan is one node in a much larger system — one that has been expanding steadily and is now accelerating.

The infrastructure began with REAL ID and the MoUs that gave the FBI access to state driver’s license databases. In December 2025, Iowa was one of four states that agreed to help the Trump administration gain access to state driver’s license data through NLETS.^[8]

The deal was part of a settlement that allowed Iowa to upload its voter rolls to the federal government for citizenship verification through SAVE,^[9] after the Secretary of State had flagged over two thousand potential non-citizen voters by cross-referencing voter rolls with DOT records — driver’s license application data that was, in many cases, years out of date.

Subsequent federal verification through SAVE confirmed only 277, roughly 12% of those flagged based on the DOT’s REAL ID records.

Meanwhile, DHS has also declared that REAL ID—the system it spent twenty years building these invasive, networked fraud-prevention and citizenship-verification systems for—is not fit for purpose.

In a December 2025 court filing, a DHS official stated that “REAL ID can be unreliable to confirm U.S. citizenship” in response to a lawsuit by an American citizen who was detained twice during immigration raids despite presenting his valid REAL ID.

What immigration enforcement now demands is being filled by facial recognition. ICE agents carry Mobile Fortify, a smartphone app that reportedly scans faces against over 200 million images across DHS, FBI, and State Department databases.

DHS does not let subjects decline to be scanned, and photos — including those of U.S. citizens — are stored for fifteen years. CBP has gone further, releasing Mobile Identify, a separate facial recognition app available to state and local police agencies deputized for immigration enforcement through 287(g) agreements.

The federal government and state DMVs spent twenty years laying the groundwork for all of this while Presidents, Governors, and members of Congress rotated in and out of service. The MoUs. The nightly scans. The databases quietly compiled from photos taken to get permission to drive — or, for non-operator ID holders, simply to prove who they are.

Iowa’s program is not an outlier. It is the foundation. It is twenty years of federally funded research, data collection, and algorithmic lineups that, by the government’s own accounting, is unreliable and serves no significant public purpose.

Iowa’s DOT does not limit itself to approving questionable permits for police surveillance cameras and waiving roadside safety standards to accommodate them. Through its Motor Vehicle Division, it also operates one of the most invasive automated surveillance programs in the state.

For 3.5 criminal charges per year — overwhelmingly misdemeanors, with no recorded convictions — and 28 investigations that led nowhere at all, every new driver’s license photo in Iowa enters the nightly automated lineup, and, along with it, the database that federal, state, and local police across the country can access on demand.

Whether you’re an immigrant or have never left Ottumwa, if IDEMIA’s computer says you’re a suspect, the government will find out everything there is to know about you—to start with.