An independent measurement · 2026

The State of AI Transparency in Europe

Europe is about to ask every AI product a simple question — can you prove you're transparent? We measured 79 of the most-used. Not one could prove it. A readiness signal, not a verdict.

The headline

What we found

0%

Of the 79 most-used AI products we measured, not one could let a third party independently verify what their AI did. 54% clearly disclose they're AI; none mark AI-generated content in a machine-readable way; and no product earned an A.

The gap that matters

Readiness by signal (R1–R6). The last bar — independent verifiability — is the gap LedgerProof exists to close.

// readiness by signal · n=79
Tells users they're talking to AI54%
Marks AI-generated content0%
AI-content labelling present28%
Independently verifiable by a third party0%

Sample of 79 of Europe's most-used AI products (20 more — including ChatGPT, Claude, Perplexity and Midjourney — block automated review and were excluded and counted), graded on observable public-surface signals against EU AI Act Article 50, on 30 June 2026. A readiness signal, not a verdict.

The grades

No product earned an A. They split almost evenly between C and D — 96% sit at C or below.

// grade distribution · n=79
A — strong readiness0%
B — mostly ready4%
C — gaps to close49%
D — significant gaps47%

Bands are A–D by design — we report readiness and a path to improve, not condemnation. The cut-offs were fixed before scanning. n=79, 30 June 2026.

What it means

Disclosure is patchy; proof is absent. A bare majority of products can say you're talking to an AI — but not one can let anyone else independently check what it did. Article 50 is, at heart, about that second thing — and it's the thing a self-written log can never deliver. The gap isn't disclosure. The gap is proof.

LedgerProof manufactures tamper-evident, independently verifiable evidence — not a verdict. This is an observed readiness signal of public-facing surfaces at a point in time; your auditor and the competent authority make the determination. We make it provable.

How we measured it

The sample frame and rubric were fixed before the data was read — that's the credibility. Each product was measured by an automated scan of its public-facing surface against the Article 50 signals — the very same free tool anyone can run on any product, including their own. We lead with aggregate findings; any product we name gets a right-of-reply and a free re-scan first. Every figure here is reproducible — which is the whole point.

Where you stand