Detection routes open to whoever holds the file
Four detection routes are published across this register. Three can be used by whoever holds a file; one is described as internal to the vendor, which makes the signal real and unverifiable from outside. As of 2026-09-22.
| Entry | Machine-readable marking | Document read |
|---|---|---|
| Amazon Nova Reel | Invisible watermark plus C2PA Content Credentials | AI service card |
| Google (SynthID) | An assistant that inspects an upload, plus a detector portal in testing | A research page |
| Sora | C2PA on every asset, standard named | Sora 2 system card |
| Veo | SynthID, with a verification platform named | Veo API documentation |
Inclusion rule. Entries whose pages name a route for establishing whether a file carries their mark. A claim that a mark exists, with no route named, is not counted here. Order. Alphabetical by the name in the first column.
1Marking and detection are different capabilities
A vendor can write a signal into every file and keep the only reader. That is not a contradiction and it is not unusual; it is simply a different product from one where a recipient can check. The register keeps the two as separate facts so that a filled file-contents cell does not imply a usable route.
Three of the four routes here are named as things somebody outside can use: a verification platform, publicly available tools, and an assistant that inspects an uploaded clip. The fourth is described as internal tools to help assess whether a video or audio was created by the vendor's products.
2Distance from the file decides what a route is worth
A publisher handed footage of unknown origin wants to establish something about the file in their hand. A route that requires the vendor's cooperation is a route through a support queue, on the vendor's timetable, with the vendor's interest in the answer. That is a different instrument from a tool a recipient runs themselves.
One of the public routes is also described as being in testing with journalists rather than generally available, which is a third position: named, public in principle, and not yet something an ordinary holder of a file can use. The register records the wording rather than grading the availability.
3What none of the four routes establishes
A detection route answers whether a signal is present. It does not answer whether a file that fails the check is genuine, because absence can mean the file was never marked, was marked by a different vendor, or was handled in a way that removed the mark. One vendor publishes precisely that caveat about its own credentials.
Nor does any route here cover another vendor's mark. Four routes, four vendors' own output, and no published cross-checking of any kind, which is worth knowing before treating any of them as a general test for generated video.
4Sources
- DetectionVideos can be verified using the SynthID verification platforma named platform
- DetectionContent Credentials can be inspected by dropping a generated video into publicly available tools such as Verifya public route
- DetectionAlso in that list: internal detection tools to help assess whether a certain video or audio was created by the vendor's productsheld inside the vendor
- DetectionGemini checks an uploaded image, video or audio clip for the watermark, and a separate SynthID Detector portal is in testing with journaliststwo published routes
Every row above is read from the vendor page named beside it, on 2026-09-22. The same values in register form are on the support table, the rule for this column is on the machine-readable marking column page, and what may fill a cell at all is on the what earns a row page. Other findings: reasons printed beside a mark, marks that get described.