ChronoVerify update, July 2026: what shipped and who it is for
Since the last update five days ago, real users have put ChronoVerify through its paces and, usefully, broken it twice. This post covers what shipped, what the field taught us, and, because several people have asked, a plain answer to the question "who is this actually for?". The backdrop has not changed: the EU AI Act's Article 50 transparency obligations apply from August 2, 2026, four weeks from now.
What shipped this week
Shareable verdict permalinks
Every verification can now mint an unlisted permalink that records the verdict, the file's SHA-256 fingerprint, and the evidence summary. The image itself is never stored. That means an analyst can cite a verdict in a report, a moderator can attach one to a ticket, and a seller can prove a listing photo checked out, all without passing the photo around. Links created on the free verifier last 90 days; links created with an API key do not expire.
A sample gallery with known ground truth
You should not have to take a verification tool's word for anything, so chronoverify.com/demo now hosts six sample images with known ground truth, one for each verdict: intact camera metadata, a geotagged photo, Content Credentials signed by an untrusted test certificate, an impossible capture time, a known pasted-region edit, and a metadata-free copy. Run them through the verifier, the API, or the browser extension and watch each verdict happen.
Raw GPS coordinates, not just a place name
When a photo carries GPS metadata, verdicts previously showed the nearest place name. Analysts want the numbers, so location now reads like "San Francisco, California, United States (37.7749, -122.4194)" everywhere a verdict renders: the verifier, permalinks, the free EXIF tool, and the extension. The coordinates come from the file itself, and the write-up on our OSINT use-case page covers how capture-time and location evidence fit an investigation workflow.
A browser extension, built and headed to the stores
Right-click any image on the web, choose Verify with ChronoVerify, and get the full verdict card in a small window. It sends only the image URL, never image bytes, and its permission list is three lines long by design. It is built and tested for Chrome and Firefox, with store submission next; a Chrome listing also covers Brave, Edge, and Opera. One behavior worth calling out: when a verdict comes back inconclusive for an image served by a platform that strips metadata, such as X, Instagram, or Reddit, the card now says exactly that and points you toward the original file instead of leaving you guessing.
Agent and developer surface
The ChronoVerify MCP server is listed in the official MCP Registry and on the largest community MCP directory, and a hosted endpoint means agents can verify images without installing anything. If you are wiring provenance checks into a pipeline, the integrations hub has copy-paste recipes for the common frameworks, and /v1/onboarding is a machine-readable manifest of the whole API.
What the field broke, and what it taught us
Two bugs were found this week by people actually using the tool, and both were fixed and deployed within hours of the report. We think the details are worth publishing.
- Every current iPhone's default photo was being rejected. Our upload gate caps decoded pixel count as decompression-bomb defense. The cap was 24 megapixels; current iPhones' default output is 5712x4284, which is 24.5 megapixels. Two percent over the line, so every default-settings iPhone photo bounced. The cap is now 40 megapixels, sized to cover every phone's default output and 33 megapixel mirrorless cameras, and the rejection message now names the actual megapixels when a file really is over.
- iPhone HDR JPGs were "unsupported". After that fix, the same reporter's iPhone JPGs still failed. Current iPhones embed an HDR gain map in their JPGs using an MPF segment, and our decoder classified those files as a separate format called MPO rather than JPEG. They are ordinary JPEG streams, and they now verify as such, including Android Ultra HDR photos. Our own test images never caught this because synthetic test JPEGs do not carry gain maps; real device files do.
The lesson we are keeping: calibration constants and format assumptions must be tested against real device output, not just synthetic files. Both failure modes now have regression tests pinned to real device shapes, and both fixes are described honestly here because a verification tool that hides its own misses has no business asking for trust.
Who ChronoVerify helps
ChronoVerify answers two questions about any image: when was this actually taken, and does the file's own evidence support what it claims to be? Here is who needs that, concretely.
- Insurance claims teams. A damage photo is only evidence if it was taken when and where the claim says. Capture-time verification catches recycled and pre-dated photos before they become payouts.
- Newsrooms and fact-checkers. Wire photos increasingly carry Content Credentials, and user-submitted images carry EXIF or nothing. One check reads both, validates signatures cryptographically, and states what the file cannot prove.
- Marketplace and resale platforms. The "recent photo" of a car, apartment, or collectible that was actually captured in 2019 is a classic fraud pattern, and it is visible in the file's own metadata when the original is checked.
- Trust and safety teams. Verdicts, hashes, and permalinks slot into moderation queues as structured triage: which reported images deserve human review first.
- KYC and onboarding reviewers. A selfie or document photo that was captured months before the session it claims to belong to is a signal worth having before approval.
- Gig-work and delivery platforms. Proof-of-service photos carry capture times. Verifying them settles disputes with evidence instead of arguments.
- OSINT analysts and investigators. Capture time, device, raw GPS coordinates, file hashes for case logs, and permalinks for citation, all from one call or one right-click. Inconclusive results on platform-stripped copies are themselves a finding: they tell you to go hunt the original.
- AI and agent builders. Article 50's transparency obligations make provenance a pipeline concern. The API returns one typed verdict object, the MCP server exposes it to agents, and a signed audit record documents each check. The reasoning is laid out in content provenance for AI systems.
- Everyone else, free. The web verifier takes an upload and answers in seconds, no account. If you are checking a dating profile photo, a rental listing, or a photo someone forwarded you, that is exactly what it is there for.
What has not changed
The boundaries hold. ChronoVerify is not an AI-image detector and does not claim to be; it validates provenance and reads the evidence a file actually carries. A clean verdict is triage, not proof a scene is real. Metadata can be stripped or forged, and the verdict language says what the file itself can and cannot support. The pipeline is deterministic: the same file returns the same verdict, and no language model scores your image. Pricing is unchanged: the verifier is free, a free API key covers 100 verifications a month with no card, and pay as you go is $0.003 per image after that.
Try it against known ground truth first, then your own photos.
Open the sample galleryCommon questions
Who is ChronoVerify for?
Anyone who has to decide whether to trust a photo: insurance claims teams, newsrooms and fact-checkers, marketplace and fraud teams, trust and safety teams, KYC and onboarding reviewers, gig-work platforms, OSINT analysts and investigators, AI and agent builders wiring verification into pipelines, and everyday people checking a listing photo or a profile picture. There is a free web verifier, a free API tier of 100 verifications per month, and pay-as-you-go pricing at $0.003 per image after that.
What does ChronoVerify actually check in a photo?
Three layers of evidence inside the file itself: C2PA Content Credentials validated cryptographically against the official C2PA and CAI trust lists, EXIF and XMP capture metadata (when the photo was taken, on what camera, and where) with internal consistency checks, and deterministic pixel forensics that look for signs of local editing. The result is one plain-language verdict with a confidence score, the file's SHA-256 fingerprint, and documented limits.
Does ChronoVerify detect AI-generated images?
No. ChronoVerify validates provenance; it does not classify whether pixels were made by a model. If an AI image carries signed Content Credentials naming the generator, validation surfaces that. If an image carries no provenance at all, the verdict is inconclusive, and it says so plainly rather than guessing.
Why did current iPhone photos fail to verify, and is that fixed?
Two separate causes, both found by real users and both fixed. First, the upload gate's decompression-bomb cap was set at 24 megapixels, and current iPhones' default output is 24.5 megapixels, so default photos were rejected as too large; the cap is now 40 megapixels. Second, current iPhone JPGs embed an HDR gain map using an MPF segment, which the decoder classified as a separate format and refused; they are now treated as the ordinary JPEGs they are, including Android Ultra HDR photos.
Can I share a verification result without sharing the image?
Yes. Every verification can mint an unlisted permalink that stores the verdict, the file's hash, and the evidence summary, never the image itself. Links created on the free verifier last 90 days; links created with an API key do not expire.
How can I test ChronoVerify before trusting it?
The sample gallery at chronoverify.com/demo hosts six images with known ground truth, one for each verdict: intact camera metadata, a geotagged photo, Content Credentials from an untrusted test signer, an impossible capture time, a known pasted-region edit, and a stripped copy. Verify them with the web verifier, the API, or by right-clicking with the browser extension.
Sources and further reading: EU AI Act Article 50; C2PA Specifications; Content Authenticity Initiative; the ChronoVerify method and limits; benchmark and calibration.