The Make Column Isn't There

Between New Year's Day and the end of April 2020, thieves in Milwaukee stole ninety-two Kias and Hyundais. Over the same four months of 2021, according to police figures reported by FOX6, they stole 1,920. Every other make of car in the city, combined, went the other way: 1,109 stolen, then 1,029.

That is one city's first look at what became the largest car-theft wave in modern American records, and then its fastest decline. Both halves of the curve were driven, to a degree nobody can measure precisely, by a single engineering decision at two affiliated automakers. And the one field that would let you see it in a public crime feed (what kind of car was taken) is missing from almost all of them.

One missing part

The part is an engine immobilizer: a chip in the key and a receiver in the car that refuse to let the engine run unless the two recognize each other. According to the Highway Loss Data Institute, immobilizers were standard on 96% of other manufacturers' U.S. vehicles by the 2015 model year. Hyundai and Kia didn't make them standard until about six years later.

Canada has required immobilizers on new vehicles since September 2007, and both companies have said their Canadian cars carry them. The U.S. standard, FMVSS 114, requires that a car not start once the key is removed. It doesn't say how. So a 2017 Elantra built for Ontario and one built for Ohio could leave the factory with different answers to the question “what happens if someone pulls off the steering column cover and turns the ignition with a USB plug?”

The method spread on TikTok and YouTube in 2021 and 2022 under the “Kia Boyz” name, which started in Milwaukee. By February 14, 2023, when the companies announced a free software upgrade for 8.3 million U.S. vehicles, the National Highway Traffic Safety Administration was attributing at least fourteen crashes and eight deaths to the trend. (Safety Research & Strategies, an advocacy group, counted considerably more from news reports. We're using the federal figure because it is the conservative one.)

The national curve

The National Insurance Crime Bureau, which builds its counts from the NCIC stolen-vehicle file rather than from police crime reports, puts the peak in 2023 and the fall since then at a pace it calls the largest in four decades.

The police-reported numbers tell the same story. The Real-Time Crime Indexshows motor vehicle theft down 22.7% in 2025 across 19,558 agencies, the steepest drop of any property category, and down another 18.8% in January through July of 2026 among the 565 agencies in its current sample. NICB counted 268,415 thefts in the first half of 2026, down 21%. (If you want the reasons those two sources don't match to the vehicle, we wrote about how the RTCI differs from the FBI's count earlier this year.)

How much of the decline is the fix? Nobody has a clean answer, and we won't pretend to. What exists is this. HLDI found in 2024 that upgraded Hyundais and Kias had theft claim frequencies 53% lower than unupgraded ones, with about 60% of eligible vehicles upgraded by mid-July of that year. In December 2025, thirty-six state attorneys general announced a settlement under which the companies will offer a free zinc-reinforced ignition cylinder protector, after the states alleged the software could be bypassed.

The other side of the ledger matters too. The Council on Criminal Justice wrote that much of the earlier increase was likely Kia and Hyundai thefts, but that “rates were already trending upward before these vehicles became popular targets.” Car theft rose in plenty of places that never had a Kia Boyz problem. A defect explains a lot of this curve. It doesn't explain all of it.

What a city file can tell you

We wanted to see what this looks like from the developer's side, inside one public feed. Dallas publishes a detailed incident file on its Socrata portal, with a free-text modus operandi field that most cities don't bother to release. Here is the query. The file has one row per involved person, so we count distinct incident numbers.

curl -s -G "https://www.dallasopendata.com/resource/qv6i-rri7.json" \
  --data-urlencode "\$select=year1, count(distinct incidentnum) as n" \
  --data-urlencode "\$where=nibrs_crime_category='MOTOR VEHICLE THEFT'" \
  --data-urlencode "\$group=year1" \
  --data-urlencode "\$order=year1"

Run on October 12, 2026, by year of occurrence:

The jump is sharp and it's late. Monthly counts sat around a thousand through 2022 and March 2023 (1,180), then climbed to 1,474 in April and 1,775 in May, and stayed high for the rest of the year. Dallas's police dashboard, as reported by Governing, counted 18,838 vehicles stolen in 2023. Our 17,825 is incidents, and one report can cover more than one car, so the two numbers shouldn't match exactly.

Two notes on reading this. First, Dallas's city systems were hit by a ransomware attack in early May 2023, the same month the count spiked. A records disruption can produce strange months, and we checked for one. The elevated level runs from April 2023 into early 2024, which looks more like a real change than a backlog flushing through, but a file can't prove that about itself. Second, the rise started well before 2023. Theft was up 42% from 2018 to 2022, which fits the CCJ point about an upward trend that predates the brand-specific wave.

The make lives in the notes

Look at the last column. In the peak year, 88 of 17,825 theft incidents mention Kia or Hyundai anywhere in the free text. That's half of one percent. Nobody in Dallas forgot what was stolen. The make went into the stolen-vehicle record that gets entered into NCIC, which is how insurers and NICB can see it. It just never made it into the file the public gets.

The MO field is officer shorthand, and it reads like it: “UNK SUSP TOOK COMPS VEHICLE W/O CONSENT.” Sometimes it's just “UUMV,” Texas's acronym for unauthorized use of a motor vehicle. The word “USB” appears in two theft narratives in nine years.

But the method leaves a trace even when the brand doesn't. Mentions of a steering column ran between 113 and 125 a year from 2018 through 2021. In 2023 there were 701. That's the one step every Kia Boyz video shows: pop the plastic cover, expose the ignition cylinder. We would not treat that column as a count of anything. Most narratives say nothing about how the car was taken, and the share that do surely varies by officer and by year. As a signal that something changed in how cars were being stolen, it is better than nothing, which is what the structured fields offer.

What it does to a trend line

Here is the practical problem. A car-theft series that ran from 2019 through 2025 contains a bump that was partly a product defect, spread by video, then partly closed by a software patch and a zinc sleeve. If you treat that series as a measure of neighborhood risk, you will score a lot of places as getting dangerous in 2022 and 2023 and safe again in 2025. Some of that is real. Some of it is whether people on the block happened to drive a 2017 Sonata.

The decline is where this bites hardest now, because it is being read as a policy result. In our view, policing and policy are getting credit for the drop in car theft that the data can't actually apportion, because the largest documented cause sits in an engineering fix that no crime feed records. That doesn't mean policing did nothing. It means you can't separate the two with the fields you have.

Andrew Wheeler, who models every city in the Real-Time Crime Index against a shared national curve, noted in a recent postthat Milwaukee and Tucson had recent vehicle-theft counts well below what his model predicted. He flagged that such drops could be data errors or could be interventions worth studying. Milwaukee, of course, is where this started. A city that had an outsized share of vulnerable cars should see an outsized fall once they are patched or scrapped. That is our hypothesis, not Wheeler's, and the public data can't test it.

We have seen this shape before. One Target store doubled San Francisco's shoplifting by changing its reporting procedure. A records migration made Phoenix look crime-free for a week. In each case the cause sat in a column the public file didn't carry. Vehicle make is the same kind of hidden variable, except it moved a national number instead of a local one.

What to do with a vehicle-theft series

1. Treat 2021 through 2025 as an episode, not a baseline.Don't compute a “normal” theft rate from those years, and don't compare a 2026 count to a 2023 peak without saying what 2023 was.
2. Separate vehicle theft from other property crime in any score. If an index blends them, a two-brand defect becomes a neighborhood safety signal. For how we think about weighting categories, see the SpotScore methodology.
3. Look for a make field, and ask for one.Check whether the agency publishes one. Most of the open incident files we ingest don't. If you're negotiating a data-sharing agreement, make and model year are cheap to include and carry no personal information.
4. Mine the narrative with a baseline.Text matches like “column” or “ignition” are only useful as a change against the same agency's earlier years. Never as a share, never across agencies.
5. Borrow from insurance data when the crime feed is silent.HLDI claim frequencies and NICB's NCIC-based counts can see make. Use them to explain a break you find, then hedge accordingly. They measure claims and stolen-vehicle entries, not crime reports.
6. Watch small areas carefully.At block level, a single resident's car stolen three times reads as a trend. We covered why in Three Burglaries Is Not a Trend.

According to NICB, the most stolen vehicle in America in 2025 was still the Hyundai Elantra, with 21,732 thefts. The fix for the vulnerable model years is a zinc collar around the ignition cylinder. It's free at the dealer, for owners who make the appointment.