Police Tech’s PR Problem
Concrete Evidence (August 17, 2026)
A day can hardly pass without some controversy over the technology available to police.
Today, everyone’s fighting about the Flock license-plate readers, which can identify stolen vehicles and tell police which cars were close to a crime scene. Not long ago (and surely again soon) the big point of contention was ShotSpotter, which alerts police to the location of likely gunshots. Soon, newer AI-based policing tools will surely garner controversy. As a new report from the National Policing Institute (NPI) warns, departments are adopting such tools far faster than safeguards are being put in place.
Using technology to bring down crime and make cops more efficient is a good thing. But the American people are especially suspicious of government surveillance, even in an age when every person you meet is carrying a video camera in their pocket. Strong safeguards against abuse are therefore an absolute must—not only to prevent abuses, but also to win the public’s trust in police tech.
Let’s take a look at that new NPI report, as well as some recent developments in the Flock debate.
A few months back, NPI and Microsoft convened numerous leaders from 13 policing agencies of varying sizes to discuss how they were using AI. Every one “had at least some form of AI presence,” NPI reports, and four-fifths had deployed at least one AI tool. The uses included writing reports, analyzing data, managing staff and schedules, and looking for warning indicators regarding officer wellness. One agency was not just using AI products, but developing them in-house—naturally, with some assistance from AI.
At the same time, two-fifths had done “no AI-specific training for any personnel.”
Police use of AI comes with a number of risks if it’s not managed carefully, some of them rather obvious. Just imagine an AI-assisted police report riddled with hallucinations, an AI data analysis that focuses police attention on the wrong people or neighborhoods, or an AI staffing program that skips some shifts.
NPI suggests a number of best practices, spotlights existing efforts to keep AI use within sensible guardrails, and urges federal guidance. Per NPI, agencies should assess where AI is being used already (as several participants “described discovering AI already embedded in workflows their agencies had not formally sanctioned”); pay close attention to what data is used, where it comes from, and who owns it; train procurement staff to closely scrutinize vendors’ claims; check with legal counsel to ensure AI use passes muster; start with low-risk internal uses and engage the community beyond that; and rigorously test systems before putting them into use.
These are good general principles. With American policing, though, one must always remember there are something like 18,000 agencies in total, operating with a great deal of autonomy. There will be bumps along the way, regardless of whatever guidance is ultimately provided by the federal government and policing groups like NPI.
Meantime, the Flock debate rages on. The issue’s central tension is that while license-plate readers operate in public places, a large enough amount of such data can allow police to track individuals’ movements in minute detail. The fact that, say, a gray Toyota Camry with the license plate ABC123 was driving down Main Street at 2:32 p.m. is not private information, but the mass collection of that information has led to some alarming abuses involving police officers tracking the whereabouts of their exes.
The company itself put out some new policies late last week. It will now require all departments to use a system that flags potentially illegitimate queries, rather than making it optional, and will lock out apparent problem users automatically until the department can review their behavior. It will also require officers to provide a “case code” (denoting what investigation they’re working on) before running a search, which departments can override in emergency situations.
Flock will also make the default data-retention period (which departments can change) just seven days. On that, though, I recommend reading this piece by Andrew Wheeler, a data scientist who’s done a lot of work with police departments. He points out that deleting old data can hamper longer-term investigations (including into homicides) but doesn’t stop a stalker from searching for his ex repeatedly.
Wheeler recommends keeping the data for a longer time, while requiring a warrant to search historical data. In his proposed system, license readers could still issue immediate alerts (“a plate reported stolen is right here right now”), and police might be able to access certain data quickly in emergencies (“what is the license plate of the car that fled the bank that was just robbed?”), but piecing together a suspect’s movements after the fact would require a warrant.
Technology can make police more efficient and solve crimes. These are enormous benefits that should dwarf any abuses in magnitude. In America, though, police have only the technologies that the public allows them to have—and rightly so.
Even a handful of news stories can turn the public against technologies they barely noticed before. Strong safeguards, therefore, don’t just prevent abuses by cops using technology. They’re what allows police to have technology at all.
From the Manhattan Institute
Don’t miss earlier Flock takes by my colleagues Rafael Mangual and Charles Lehman.
Other Work of Note
Just posted yesterday: A big new working paper on Flock from Scott M. Mourtgos and Ian T. Adams (who’ve both written for City Journal). They find a roughly one-tenth drop in motor vehicle theft, and a possible improvement in clearances as well, when agencies adopt the technology. And don’t miss my colleague Charles Lehman’s website compiling cases where Flock has helped stop crimes.
Some regularly updated data about the effects of AI on the economy. Also findings from OpenAI about how firms are using the technology, a study of AI use in applications for federal funds, and a collection of articles about AI and the law (including its use in hiring and college admissions, where discrimination concerns are paramount).
Oh, and faced with two options, AIs have a bias toward the one they saw first, as David Rozado also found in this MI research. They can also play Civilization V!
What do the data actually say about grocery prices over time?
What causes the lower average mental health of sexual minorities?
Are gender gaps in personality disappearing among the young?
Some polling data about transgender athletes in women’s sports.
Tracking men’s marijuana use. (It’s up.) And some clarifications on what it means to “decriminalize” drugs.
The expansion of higher ed in the U.K. “led to the selection into college of progressively less talented students from advantaged backgrounds.”
When a union narrowly wins an election, it boosts unions’ chances in other elections in the same state and industry.
Immigrant earnings assimilation since 1981.
How property taxes change how housing is allocated between young adults and seniors.
“Opportunity Zones” were in large part designated in places where new projects were already being planned. And how zoning rules affect Boston housing.
Is there actually more walking in places considered more “walkable”?
On the relationship between genetics, educational attainment, and teachers’ encouragement of some kids more than others.
Exposure to “untrustworthy sources” on Facebook is highly concentrated among a subset of users, and reducing it doesn’t seem to change attitudes or beliefs, this Science Advances study finds.
How close are American schools to environmental hazards? And which types of housing are worst for the environment?
Studies of environmental topics seem especially heavy on subjective language.
Trends in “implicit bias” measures in cities. If you missed it, I had a skeptical column on the subject a while back.
Support for free speech is strongest at both the right and left tails of the political spectrum.
Interesting results on “no-split” rules for redistricting (meaning cities and counties can’t be split up).
The U.S. seems headed for another record-low murder rate.
Why do people lose touch with their probation officers?
Is gun violence exposure a big driver of racial health disparities?
The comments sections of political YouTube videos seem less polarized than the videos themselves.
Thank you Robert, appreciate the positive review of my post.