ChronoVerify update, July 18, 2026: an Article 50 marking checker and richer reports
One day since the last update, and this one is about access: a new free tool for the EU AI Act deadline that was then fifteen days out, and signed reports that finally show you the image they describe.
What shipped
A free Article 50 marking checker
The EU AI Act's transparency obligations apply from August 2, 2026, and Article 50(2) expects providers of generative AI systems to mark their outputs in a machine-readable format. The new Article 50 marking checker examines one image for exactly that marking: a C2PA Content Credential, cryptographically validated against the official C2PA and CAI trust lists, and the IPTC digital source type declaration (such as trainedAlgorithmicMedia, the code for content created by generative AI), whether it appears inside a signed manifest or in plain XMP metadata. Free, no signup, nothing retained.
This is a marking check, not an AI detector: absence of marking proves nothing, because marking is commonly stripped by resizing, re-saving, screenshots, and most social platforms. A declaration in plain XMP without a Content Credential is editable metadata, not proof. And under the AI Omnibus package adopted in June 2026, generative systems already on the market have until December 2, 2026 to meet the marking requirement, so check the current state of the law rather than treating August 2 as a universal cliff.
Signed reports now show the image, and explain every verdict
Two things a first-time reader of a ChronoVerify report asked for, reasonably: what image is this about, and what do the verdict words mean? Reports now open with a colored verdict banner, embed a rendering of the analyzed image labeled visual reference only, and include a verdict scale table that defines all five verdicts in one sentence each, with the returned verdict highlighted. The rendering is for orientation; the evidence remains the original file, identified by the SHA-256 fingerprint the Ed25519 signature covers, with the RFC 3161 trusted timestamp from yesterday's update anchoring the record in time.
The verifier now previews your photo
Drop a photo on the free verifier and you now see a small preview beside the verdict. The preview is drawn by your own browser from the file you selected; it is never uploaded for display, never stored, and never served back by us. The check itself still processes the image in memory and retains nothing, and shareable verdict links still contain the verdict only, never the photo.
Smaller changes worth a line each
- An API link in every nav. Developers looking for the API found only a Docs link; every page now carries both.
- chronoverify-mcp 0.1.6. The MCP server's signed-report tool now describes the embedded RFC 3161 timestamp token, and the hosted and npm versions report the same number.
- Error messages that name the real fix. A customer refused a report is now told reports draw from the credit balance on their plan, instead of a misleading out-of-quota message.
- More verification capacity. The concurrency ceiling was raised after live load testing, so bursts that previously shed a request or two now serve cleanly.
What has not changed
The boundaries hold. ChronoVerify validates provenance and reads the evidence a file actually carries; it is not an AI-image detector. A clean verdict is triage, not proof a scene is real. The pipeline is deterministic: the same file returns the same verdict, and no language model scores your image. The free verifier stays free, a free API key covers 100 verifications a month with no card, and pay as you go stays at $0.003 per image with a $5 minimum top-up.
The current ladder: free, pay as you go, and Scale. Keys appear on screen right after checkout.
See pricingCommon questions
Which plan should I pick?
Free covers 100 verifications a month with no card, and the public web verifier is unlimited with no key. Pay as you go is $0.003 an image drawn from a prepaid balance, $5 minimum top-up, with signed PDF audit reports at $2.00 each from the same balance. Scale is $299 a month for 200,000 verifications with signed, timestamped PDF audit reports included in that quota.
Does the Article 50 marking checker detect AI images?
No. It checks one file for the machine-readable marking Article 50 expects providers to apply: a validated C2PA Content Credential and the IPTC digital source type declaration. Absence of marking is not evidence an image is or is not AI-generated, and the page says so.
Is the image shown in the signed report the evidence?
No. The embedded picture is a re-encoded rendering labeled visual reference only, so a reader can see what the report is about. The evidence is the original file, identified by the SHA-256 fingerprint the Ed25519 signature covers. Verify any copy by hashing it and comparing.
Does the verifier preview upload my photo somewhere?
No. The preview is drawn by your own browser from the file you selected, using a local object URL. The server processes the image in memory for the check and retains nothing, and it never serves the image back. Shareable verdict links contain the verdict only, never the photo.
When does Article 50 apply?
The transparency obligations apply from August 2, 2026. Under the AI Omnibus package adopted in June 2026, generative AI systems already on the market before that date have until December 2, 2026 to meet the machine-readable marking requirement, and the European Commission's guidelines have since been adopted, on July 20, 2026, as guidance rather than binding law.
Is ChronoVerify C2PA conformant?
Yes. ChronoVerify is a C2PA Conformant Validator on the C2PA Conforming Products List, record 019f8a20-6452-7a43-b11b-59d0b0e4a84a, covering validation of JPEG, PNG, WebP and AVIF under C2PA specification 2.2. The list is public, so you can check the record yourself. It covers validation rather than generation: ChronoVerify reads and validates Content Credentials, it does not sign them.
Sources and further reading: EU AI Act Article 50; IPTC digital source type vocabulary; C2PA Specifications; ChronoVerify pricing; the method and limits.