Correction, 14 August 2026. As first published, this study said the earlier wording of a United States recall is “overwritten with no public history”. That is wrong. NHTSA publishes amended notification documents as separate PDFs, so an earlier version does survive in the document set. What is overwritten is the structured, machine-readable record: the feed that aggregators, dashboards and AI assistants actually read. The claim has been narrowed accordingly. We restate corrections rather than editing silently.
A recall almost never stays the size it was announced. More cars get added, the remedy gets rewritten, the instruction to stop driving appears or disappears. What nobody has written down is that the world’s safety authorities handle that growth in two fundamentally different ways, and both of them hide the correction from the public.
We went looking for a simple thing: which vehicle-safety authorities revise a recall in place after publishing it, and which file a second recall instead? The answer turns out to determine whether a change can be tracked at all.
What we did
Between 11 and 14 August 2026 we read the published recall records of nine jurisdictions directly from the authority’s own website or API, not from press coverage, and not from aggregators. Where an authority publishes a procedural rulebook, we read that too. For the three countries we already monitor continuously, we compared successive readings of the same recall to observe revision behaviour rather than infer it.
This is a study of publishing practice, not of safety performance. Nothing here says one authority protects drivers better than another.
The finding: two ways of handling a recall that grows
| Authority | Behaviour when scope grows | What the public can still see |
|---|---|---|
| United States NHTSA |
Revises in place. Affected-vehicle counts, remedy text and consequence statements are all edited after filing. | The machine-readable record carries only the current values. Amended notification documents are published as separate PDFs, so a history exists in the document set, but it is not in the structured feed every downstream system reads. |
| Canada Transport Canada |
Revises in place, and moves a last-updated date when it does. | The date reveals that something changed. It does not reveal what. |
| United Kingdom DVSA |
Monitored continuously here, but our recording window is not yet long enough to characterise its habit honestly. | To be established. We would rather leave this blank than guess. |
| China SAMR |
Does not revise. Expansions are filed as an entirely new recall number. | Both records survive, but nothing links them, so the pattern is invisible. |
| Japan MLIT |
Does not revise. Its own filing rulebook contains no provision for amending a notification; expansions get a new number. | Both survive, unlinked, and only as PDF press releases. |
Two different ways of hiding the same thing
Where a recall is edited in place, the correction is invisible because the structured record now shows only the new values. Where a second recall is filed instead, the correction is invisible because nothing joins the two records together.
The practical effect is identical. A reader who saw the original announcement (or a journalist who reported the original number) has no way to discover that the story moved, unless somebody was watching at the time and kept a copy.
This is the gap our vehicle recall change register exists to close. It reads the US, Canadian and UK registers on a schedule and records every revision it observes, with the date. For figures it keeps the earlier value; for rewritten wording it currently records the change and the current text, and says so plainly.
What each authority actually publishes
| Jurisdiction | Machine-readable? | Amendment date? | Roughly how often |
|---|---|---|---|
| United States: NHTSA | Yes, JSON API | No, but fields change | ~17–19 campaigns a week |
| Canada: Transport Canada | Yes | Yes, last-update date | Continuous |
| United Kingdom: DVSA | Yes | No | Continuous |
| China: SAMR | Server-rendered HTML, no API | No | 190 automotive campaigns in 2025, covering 6.85m vehicles |
| Japan: MLIT | No: PDF press releases only | No | 358 notifications in FY2025, 4.01m vehicles |
| India: SIAM | HTML table, no identifiers | No | Publisher is a trade association; the listing has not been updated since February 2022 |
| South Korea: MOLIT | Documented REST API | Not established | Not established: the API requires a registered key and was not reachable from our network |
| Brazil: Senacon | Catalogued as CSV/XLSX/JSON | Not established | Published under a NoDerivatives licence, which forbids the kind of reuse a change register requires |
| EU: Safety Gate | Yes | No amendment field at all | Not usable for tracking revisions |
Why this matters beyond record-keeping
Three consequences follow from the table above.
Recall totals are not comparable across countries. A jurisdiction that files expansions as new numbers will report more campaigns for the same underlying defect than one that edits the original. Any league table of “which country has the most recalls” is measuring filing convention as much as safety.
The number you read is a snapshot, not a fact. In three days of watching the United States register in August 2026 we recorded six remedy rewrites, one summary rewrite and one revision of the affected-vehicle count: on that last one, downwards, from 25 vehicles to 23.
Tracking is only possible where somebody keeps the earlier copy. No authority in this study exposes a version history in its machine-readable feed. NHTSA does publish amended notification documents as individual PDFs, so the history is not destroyed, but reconstructing it means parsing the document set rather than reading the data.
What we could not establish
Publishing the limits matters as much as publishing the findings.
- The United Kingdom’s revision habit. Our continuous recording began on 12 August 2026. That is not long enough to characterise DVSA, so we have not.
- South Korea. A genuine recall API exists, but every Korean recall host timed out from our network while the parent ministry’s site loaded normally: consistent with geographic restriction rather than an outage. We could not read a single field, so we claim nothing about its contents.
- Brazil. The dataset is catalogued but the portal API returned 401 without a registered key. The licensing question is the more important one and is unresolved.
- Whether in-place revisions are ever announced. We observed that revisions happen. We did not establish whether any authority notifies previously-notified owners when the remedy text changes. That is the obvious next study.
Sources
Every figure above was read from the publisher’s own document between 11 and 14 August 2026.
- NHTSA recall API:
api.nhtsa.gov/recalls, read continuously since 11 August 2026 - Transport Canada and DVSA recall registers, read continuously since 12 August 2026
- SAMR defective-product recall announcements,
samr.gov.cn/zlfzj/qxcpzh/, and the 2025 annual notice - MLIT recall index,
mlit.go.jp/jidosha/recall.html; FY2025 statistics; and 「リコールの届出等に関する取扱要領」, the filing rulebook - SIAM voluntary recall portal,
siam.in - MOLIT vehicle recall API listing,
data.go.krdataset 15089863 - Senacon recall dataset,
dados.gov.br, and its licence statement ongov.br/mj
Questions this study answers
Do safety authorities change a vehicle recall after it has been announced?
Yes. In the United States and Canada, published recalls are revised in place: the affected-vehicle count, the remedy text and the consequence statement are all edited after filing. Watching the United States register for three days in August 2026, we recorded six remedy rewrites, one summary rewrite and one change to the affected-vehicle count.
Which countries revise a recall in place, and which file a new one?
The United States (NHTSA) and Canada (Transport Canada) revise the published record in place. China (SAMR) and Japan (MLIT) do not revise at all: when the scope of a defect grows they file an entirely new recall number. Japan’s own filing rulebook contains no provision for amending a notification.
Does any authority keep the earlier version of a recall?
No authority in this study publishes a version history of its own recalls. Where a recall is edited in place the earlier wording is overwritten. Where a second recall is filed instead, both records survive but nothing links them together.
Why can’t recall totals be compared between countries?
Because filing convention differs. A jurisdiction that files expansions as new recall numbers will report more campaigns for the same underlying defect than one that edits the original record. Any ranking of which country issues the most recalls is measuring paperwork as much as safety.
How can I find out whether a specific recall was revised?
Only if somebody was reading the register at the time and kept a copy. Our vehicle recall change register reads the United States, Canadian and United Kingdom registers on a schedule and keeps every earlier version, so the wording a recall carried before it was revised is still available.
Related studies
- The over-the-air recall the paperwork misses: the same problem inside a single authority, NHTSA’s structured over-the-air field counted three recalls in 2025, while an independent count of the remedy text found forty-two.
- Most “stop driving” recalls are tiny: 2.4 million vehicles under a do-not-drive or park-outside instruction, but a median campaign of 629 cars, and only five of sixty fixable by software.
How to cite this study
CarsMultiverse (2026). How the world’s recall authorities handle a recall that grows. Published 14 August 2026. https://carsmultiverse.com/research/how-recall-authorities-handle-revisions/
The figures were read on the date stated and are held with the wording they carried at the time. If they are restated, the change is noted on the page rather than made silently, so a citation to this version remains checkable.
Underlying records: the changes as JSON · RSS feed · the register itself.
Use this
This study is free to quote and republish with credit to CarsMultiverse and a link to this page. If you are reporting on a specific recall and want to know whether its published details have changed since it was announced, the change register holds the earlier versions, and what changed this week is the running summary.
Published 14 August 2026. This page will be updated as the recording window lengthens; material changes will be noted here and on our corrections page.