Direct questions

28. Can AI detect fake classic car provenance documents?

The Vintage Vault research deskSection 01Page 28 of 30Last updated 17 Nov 2026

AI can help organise and question records, but its output is not authentication. NIST identifies confidently stated false content—confabulation—as a generative-AI risk, which is directly relevant when a system summarises an incomplete or contradictory history file. (nvlpubs.nist.gov)

Useful screening tasks include:

  • Extracting dates, identifiers, mileage entries and invoice references.
  • Finding inconsistent names, currencies, descriptions or chronology.
  • Comparing repeated advertisements with earlier versions.
  • Producing a list of claims requiring issuer confirmation.
  • Linking every extracted claim to the original page or image.

The desk’s proposed safeguards are straightforward: preserve original files; retain source references; label extracted text separately from interpretation; require human review of important findings; and confirm critical records with their issuers or appropriate specialists.

A persuasive upload may still require examination of the original document and the car. Conversely, an OCR error or unusual historical format should not be labelled fraud without investigation.

Agree confidentiality, access, retention and deletion terms before uploading sensitive records. Do not submit unredacted owner identities, financial information or complete private archives to an unapproved tool simply to accelerate an initial screen.

Sources cited on this page

  1. nvlpubs.nist.govnvlpubs.nist.gov

Explore the research