A visual check, does the letterhead look right, does the signature look plausible, was never rigorous, but it was cheap and usually good enough to catch obvious fraud. AI image and document generation has closed that gap: a convincing forgery no longer requires skill or time. The response isn’t a better visual check, it’s not relying on appearance at all, and checking the cryptographic proof instead.
Forging a convincing signed document, matching a company’s letterhead, formatting, signature style, and plausible metadata, used to take real skill or real effort. That cost was itself a filter: most fraud attempts were sloppy enough to catch on sight, so a human glance was a reasonable, if informal, first line of defense before a document was acted on.
Generative AI tools now produce convincing document layouts, plausible signatures, and consistent formatting with minimal skill and effort, on demand. The cost that used to make forgery self-limiting has largely disappeared. A document that looks indistinguishable from genuine at a glance no longer implies it took real effort or skill to fake, which means it no longer implies it’s genuine.
High-stakes contracts already tend to get real scrutiny. The actual exposure is in the high-volume, routine documents, invoices, delivery confirmations, signed amendments, vendor paperwork, that get a quick glance and are acted on, precisely because they haven’t historically warranted more. Those are exactly the categories where an increasingly cheap forgery has the most room to do damage before anyone looks closely.
A visual check evaluates appearance, which is now cheap to fake convincingly. Independent cryptographic validation evaluates something else entirely: whether an embedded seal, checked against a public trust list, actually chains to a real, supervised certificate, and whether the file has been altered since that seal was applied. See how to validate a signed PDF independently. Convincing appearance and a valid cryptographic chain are unrelated properties; getting better at faking one does nothing to fake the other.
It doesn’t mean every routine document suddenly needs a full audit. It means the old informal filter, does this look right, is no longer a safe enough signal on its own for anything where being wrong is costly. The documents worth running an actual check on are exactly the ones that used to feel too routine to bother.