TrustVerify: detecting fakes is the wrong goal
Deepfake fraud took around €860 million in Europe last year, and attacks have risen more than twenty-fold since 2022. Chasing the fakes is a losing race. Certifying what is real is a different problem.
The numbers in this field move faster than the defences. Reported deepfake attacks are up more than two thousand per cent since 2022. Roughly six in ten organisations have no budget allocated to the problem, and four in five have no response protocol for the day it happens to them. Meanwhile the tooling to produce a convincing forgery keeps getting cheaper and the tooling to spot one keeps getting more specialised.
Detection alone is structurally a losing position. Every detector is trained on the forgeries that already exist, and every new generation of forgery is produced by someone with access to the detectors. Playing that game means permanently defending against something that improves faster than you do.
TrustVerify inverts the question. Rather than trying to prove a piece of content is fake, it establishes and records what is verified as real — which is a claim that can be made once, at the point of capture or publication, and checked by anyone afterwards. That is the same structural move VaultTrace makes for AI computation: stop asking whether the evidence looks trustworthy and start producing evidence that can be checked.
The regulatory clock is the reason this is urgent rather than interesting. European rules now require visible and machine-readable labelling of AI-generated content, with penalties at the top of the scale. Financial regulators have told institutions to adapt their anti-money-laundering systems to fabricated identities and forged documents. For a European organisation, this stopped being an optional investment and became an obligation with a deadline attached.
TrustVerify is being built inside the NovaReg regulatory line, natively against European requirements rather than adapted to them afterwards — which is the ordinary reason American products arrive in this market late and awkwardly.
Where it stands: architecture and blueprint. There is nothing to buy and nothing to download. The figures above are drawn from published industry reporting rather than from our own measurement, and we would rather cite them that way than present someone else’s research as our finding.