NHTSA's Own "Over-the-Air" Recall Flag Has Been Switched Off Since May 2025 - 64 Recalls on 11.5 Million Vehicles Say One Thing in Writing and Another in the Data

The short answer

Key facts

• N = 13,689 distinct US vehicle recall campaigns in NHTSA's Part 573 flat file, retrieved 2026-10-19. 159 of them state in their official remedy text that the fix is delivered over the air.
• NHTSA's public recall API disagrees with that text on 69 of the 159 (43.40%).
• The disagreement is not scattered. It is a clean break: 0% wrong through 2023, 6.1% in 2024, 89.5% in 2025, 100% in 2026.
• The last correctly flagged campaign was filed 8 May 2025 (25V282000, Volvo). Every one of the 64 filed on or since 9 May 2025 is flagged false.
• Those 64 come from 20 different Part 573 filers; the biggest, Ford, is 16 campaigns and 8,002,299 vehicles.
• In a random sample of 150 recent vehicle recalls whose remedy text does not mention over-the-air, the flag was true on 0 - so since the break it is not being set on anything.

Has NHTSA stopped flagging over-the-air recalls?

On the evidence, yes - some time around 9 May 2025, and without the field being removed. The flag is still returned on every campaign, still true on older recalls, and still false on everything new.

The 2020-2023 block is the control that makes the rest readable. If the flag had always been unreliable, this would be a story about a field nobody ever finished. It was set correctly on 53 of 53 campaigns over those four years, so the mechanism worked, and then stopped working.

Which cars are affected?

These are the six largest campaigns in the 64 that NHTSA's data no longer flags as over-the-air.

By manufacturer, the 64 campaigns break down as Ford 16 (8,002,299 vehicles), General Motors 8 (221,608), Hyundai 6 (293,259), Lucid 5 (66,259), Rivian 3 (157,866), Kia 3 (125,526), BMW 3 (108,241), and 16 further filers with one or two each, down to Polestar with a single campaign covering 2 vehicles.

Why does this matter if you are buying a used car?

Because an over-the-air recall is the one kind of recall that can be repaired with no dealer visit and therefore no service record.

A conventional recall leaves a trace you can look for: the car went to a franchised dealer, and that visit tends to appear in service history. An over-the-air fix can install itself overnight while the car sits on a driveway. Nobody signs anything. So for this growing class of recalls, neither completion nor non-completion is reliably written down in the places a buyer can check - and the one machine-readable signal that would at least tell you this is that kind of recall is currently not being set.

That cuts both ways, and the honest version includes both:

• A recall showing as open on a car you are considering may already be fixed, if the owner's car was online and took the update.
• A recall that is genuinely still open can stay open indefinitely if the car was never connected, and no missed dealer appointment will reveal it.

What a VIN check can and cannot tell you here

It can tell you that a recall exists against a specific vehicle, alongside what a Zilocar report is built for: accident and damage records including airbag deployment, odometer readings and rollback indicators, salvage-auction and theft records, ownership history and sales-listing history.

It cannot tell you whether the repair was done. Remedy status - whether a dealer performed the work, or whether a software update was pushed and accepted - is not part of a vehicle history record. Nor are open NHTSA investigations, which are a separate filing. For remedy status, use NHTSA's own VIN lookup at nhtsa.gov/recalls and ask a franchised dealer to read the vehicle's campaign status directly. We cede that ground deliberately: a history report that claimed to prove a software recall had installed would be making it up.

The checks we ran against our own finding

A finding this clean deserves the test most likely to break it, so here are four, with results either way.

Is it an artefact of the endpoint we queried? No. We re-pulled six of the mis-flagged campaigns from NHTSA's other public endpoint, recallsByVehicle, which serves the consumer-facing lookup rather than a campaign query. All six returned overTheAirUpdate: false there too. That pull also independently reproduced the break date: the newest campaign still flagged true on any of those vehicles was 25V282000, filed 8 May 2025 - the same campaign our main analysis identifies as the last correct one.

Is the flag simply being filled in late? The evidence is against it. Lucid filed 24V495000 on 28 June 2024 (flagged false) and 24V497000 three days later (flagged true); Hyundai filed 24V529000 on 11 July 2024 (false) and 24V583000 on 7 August 2024 (true). Same filers, same weeks, opposite flags, and the false ones have had over two years to be corrected.

Is it one manufacturer's filing habit? No. The 64 campaigns span 20 distinct Part 573 filers, the top filer holds 25.0% and the top three 46.9%. It is also the wrong place to look: the manufacturers' remedy text is correct in all 64 cases. The field that is wrong is the agency's.

Did the premise we opened on survive? No, and we are reporting that because it is the most useful part. This analysis began from a spot-check of 26 recent camera recalls which all carried overTheAirUpdate: false, suggesting the field had simply never worked. Run across the whole file, that is wrong: the flag is correct on 90 of the 159 campaigns, including 53 of 53 from 2020 to 2023. Had we published the spot-check, the first person to query a 2022 recall would have falsified it.

What this does not show

• It does not show that any vehicle is unrepaired. The flag describes how a remedy is delivered, not whether it was completed. We have no completion data and make no claim about it.
• It does not show intent. We do not know whether the field was deprecated, broken in a pipeline change, or deprioritised, and we did not ask NHTSA before publishing. The dates are what the data supports; the reason is not.
• It does not cover every OTA recall. The universe is campaigns whose remedy text says over-the-air. A recall fixed remotely but described in other words is invisible to this method, so 159 is a floor, not a census.
• 2026 is partial, closing at the file's newest report-received date of 8 October 2026.
• There is no state-level cut. The recall flat file carries no state field.

Method

Sources. Two public NHTSA datasets, matched on campaign number.

1. The ODI Part 573 recall flat file, FLAT_RCL_POST_2010.txt, from https://static.nhtsa.gov/odi/ffdd/rcl/FLAT_RCL_POST_2010.zip. Downloaded 2026-10-19 at 07:14 UTC; server Last-Modified: Mon, 19 Oct 2026 07:04:09 GMT; 246,109 rows; archive extracted without error.
2. The public recall API, https://api.nhtsa.gov/recalls/campaignNumber?campaignNumber=<CAMPNO>, queried once per campaign on 2026-10-19 between 07:16 and 07:19 UTC. All 160 queried campaigns returned results; 0 requests failed or were retried to exhaustion.

Universe. Rows with RCLTYPECD = V (vehicle), collapsed to one record per CAMPNO because the flat file carries one row per make/model/year: 13,689 distinct campaigns. Note that V means vehicle, not car - it includes motorcycles, RVs and heavy trucks, and three of the 64 mis-flagged campaigns are RV, motorcycle or heavy-truck filers (Grand Design, Zero Motorcycles, Mack).

Filter. A campaign enters the comparison universe when its remedy text - CORRECTIVE_ACTION, field 22 - matches over-the-air, over the air followed by update/software/repair/remedy, (OTA), or OTA as a standalone word, case-insensitive. That returned 160 campaigns. Every match was read. One was a false positive - Crossroads RV campaign 16V409000, whose remedy fits trim to "cover the air gap" behind a range - and was removed, giving N = 159.

Comparison. Each of the 159 campaign numbers was queried against the API and its overTheAirUpdate boolean recorded. The API returns one row per make/model/year; the flag was single-valued across all rows of every campaign, so agreement is counted once per campaign, never per row. A campaign is counted as disagreeing when its remedy text says over-the-air and the flag is false.

Converse test. To establish that the flag is not merely being used to mean something else, 150 campaigns were drawn at random (seed 19) from the 1,179 vehicle campaigns filed on or after 9 May 2025 whose remedy text does not mention over-the-air. The flag was true on 0 of them.

Known limits. Text matching defines the universe, so a remotely-delivered remedy described in other language is not counted - 159 is a lower bound. The flag's intended definition is not documented in the API response itself, so "correct" here means "agrees with the campaign's own remedy text", which is the only standard available from published data. We did not contact NHTSA before publishing. Figures for 2026 are partial.

Frozen dataset. All 159 campaigns, one row each, with the filer, makes, model years, component, unit count, the API flag, whether it agrees, and an excerpt of the remedy text: ota flag vs remedy text, 2026-10-19 (CSV)

Sources

• NHTSA ODI Part 573 recall flat file (public domain): https://static.nhtsa.gov/odi/ffdd/rcl/FLAT\_RCL\_POST\_2010.zip - retrieved 2026-10-19 07:14 UTC, Last-Modified 2026-10-19 07:04:09 GMT
• NHTSA public recall API, campaign endpoint: https://api.nhtsa.gov/recalls/campaignNumber - queried 2026-10-19
• NHTSA public recall API, vehicle endpoint (used for the independent cross-check): https://api.nhtsa.gov/recalls/recallsByVehicle - queried 2026-10-19
• NHTSA recall and VIN lookup for remedy status (cited by address; not machine-retrieved for this analysis): https://www.nhtsa.gov/recalls
• Our frozen dataset for this article: https://zilocar-news.github.io/zilocar-rss/data/nhtsa-over-the-air-recall-flag-stopped-vin-check-2026-2026-10-19.csv

Last verified: 2026-10-19