Key Takeaways
- Utah law permits facial recognition only for felony investigations, violent crimes, threats to human life, and identifying deceased, incapacitated, or at‑risk individuals.
- All requests must be routed through the Utah Department of Public Safety, which conducts a two‑person manual review after automated biometric analysis.
- Between July 1 2024 and June 30 2025, agencies submitted 1,191 requests; 706 (≈59%) yielded a probable match.
- The Brad Johnston case illustrated how a false match can lead to felony charges, public distress, and eventual case dismissal despite the technology’s “lead‑only” status.
- Officials stress that human judgment remains essential, but confidence thresholds are flexible and decisions rely heavily on reviewer discretion.
- Police chiefs acknowledge rapid AI advances and urge policies that serve both effective policing and public trust.
- Ongoing scrutiny highlights the need for transparent oversight, regular audits, and clear limits to prevent misuse while preserving legitimate investigative benefits.
Overview of Utah’s Facial Recognition Law
Utah has positioned itself as one of the nation’s most restrictive states regarding law‑enforcement use of facial recognition technology. Statute allows the tool only in narrowly defined circumstances: felony investigations, violent crimes, imminent threats to human life, and efforts to identify deceased, incapacitated, or otherwise at‑risk individuals. The law expressly prohibits routine surveillance, traffic stops, or low‑level misdemeanor inquiries. By confining the technology to serious public‑safety matters, legislators aimed to balance crime‑fighting capabilities with privacy protections. Any request that falls outside these categories must be denied, and agencies are required to document the justification for each use in a centralized log accessible to oversight bodies.
How Facial Recognition Requests Are Processed
When a Utah law‑enforcement agency wishes to employ facial recognition, it must submit the request to the Utah Department of Public Safety (DPS). DPS serves as the sole gatekeeper, ensuring compliance with statutory limits before any analysis begins. Upon receipt, the system extracts biometric facial features from the submitted image—typically a still from surveillance video—and compares them against the state’s driver‑license photograph database. Automated algorithms generate a similarity score, but the process does not end there. Two trained DPS analysts independently review each comparison, looking for contextual cues that the algorithm might miss, such as lighting variations, facial obstructions, or demographic biases. Their joint determination yields one of two outcomes: a “possible match” or “no result.” The analysts’ involvement ends once the finding is returned to the requesting agency, which must then treat the lead as investigative rather than conclusive evidence.
Statistical Overview of Usage (2024‑2025)
Data released by DPS for the fiscal year July 1 2024 – June 30 2025 reveal a growing reliance on the technology. Over that period, Utah agencies filed 1,191 facial recognition requests. Of those, 706 resulted in a probable match, translating to a roughly 59 % success rate. The remaining 485 requests produced no match, either because the subject’s image was not in the driver‑license database or because the similarity score fell below the threshold deemed worthy of human review. The upward trend in request volume—Jensen noted a steady increase over the past five years—reflects both expanding familiarity with the tool and the perceived value of rapid lead generation in serious cases. Nonetheless, the modest match rate underscores that facial recognition remains an investigative aid rather than a definitive identification method.
The Brad Johnston Case: A Cautionary Tale
The technology’s limits came under public scrutiny in the case of Brad Johnston, who was charged with felony vandalism after a facial recognition match linked him to surveillance footage from inside an Uber vehicle. Johnston steadfastly maintained his innocence, describing the experience as “just terrifying.” The match originated from a still image captured during the ride, which investigators submitted to DPS for comparison against driver‑license photos. Although the automated system flagged a possible match, the ensuing months of court proceedings revealed inconsistencies that ultimately led prosecutors to dismiss the charge. Johnston’s ordeal highlighted how a false positive—whether due to algorithmic error, poor image quality, or human misinterpretation—can trigger serious legal consequences, erode public trust, and waste judicial resources. The case also demonstrated that, even when a match is labeled merely an “investigative lead,” its presence can shape charging decisions and pretrial detention.
Human-in-the-Loop Verification and Its Limits
Tanner Jensen, chief of investigations for DPS, emphasized that the department’s protocol relies heavily on human oversight. After the algorithm generates a similarity percentage, two analysts manually scrutinize the pairings. Jensen noted that scores below 90 % do not automatically disqualify a candidate, nor do scores above that threshold guarantee confidence. “You may get a percentage below 90%, but that’s not to indicate that that’s not the individual,” he explained. “Or you may get a percentage that’s above 90% and we still don’t feel confident that that would be the individual. It really comes down to the human‑in‑the‑loop aspect.” This approach acknowledges that facial recognition algorithms can be brittle—susceptible to variations in pose, expression, occlusion, and demographic bias—making contextual judgment indispensable. However, the reliance on human reviewers also introduces subjectivity; differing training, fatigue, or implicit biases can affect outcomes, underscoring the need for standardized rubrics and regular calibration exercises.
Perspectives from Law Enforcement Leadership
Retired Salt Lake City Police Chief Chris Burbank observed that police agencies have historically embraced new technologies swiftly, likening the current facial‑recognition surge to the adoption of body‑worn cameras a decade ago. “The technology is just moving so fast and furious,” Burbank said, noting the growing power of AI to sift through massive databases in seconds. He cautioned, however, that speed must not outpace policy. “We need to ensure, again, is this policy sound for the public or is it just good for policing?” Burbank’s comment reflects a broader call for oversight mechanisms that evaluate not only operational efficacy but also civil‑rights impacts. He advocated for periodic audits, community‑engagement forums, and transparent reporting of match rates, false positives, and downstream judicial outcomes.
Balancing Public Safety and Civil Liberties
The tension between effective crime‑fighting and privacy protection lies at the heart of Utah’s facial‑recognition framework. By limiting use to serious offenses and mandating a centralized, reviewed process, the state attempts to mitigate the risk of indiscriminate surveillance. Yet incidents like the Johnston case reveal that even safeguarded systems can produce harmful errors when image quality is suboptimal or when human reviewers over‑rely on algorithmic suggestions. Civil‑liberties groups argue that the mere existence of a searchable biometric database poses a chilling effect on free association and dissent, especially if future expansions were to include non‑driver‑license sources such as social media or public‑camera feeds. Policymakers therefore face the ongoing challenge of refining thresholds, improving algorithmic fairness, and ensuring that any expansion of use remains subject to rigorous legislative scrutiny and public accountability.
Future Outlook and Policy Considerations
Looking ahead, Utah’s law‑enforcement leaders anticipate continued growth in facial‑recognition utilization as AI models become more accurate and accessible. To harness benefits while curbing risks, several policy enhancements have been proposed: (1) establishing a mandatory error‑rate disclosure for each request, requiring agencies to report both match and non‑match outcomes alongside contextual notes; (2) implementing periodic independent audits of DPS’s review process to detect systemic bias or procedural drift; (3) tightening the definition of “violent crime” and “threat to human life” to prevent mission creep; and (4) funding research into algorithmic transparency tools that can highlight why a particular match was flagged, thereby supporting more informed human judgment. By embedding these safeguards into statute and practice, Utah could serve as a model for other states seeking to leverage facial‑recognition technology responsibly—protecting communities without compromising the fundamental rights that underpin democratic policing.