Detection
The OpenAI provenance check
Our API sends the image file alone to OpenAI's content provenance endpoint, which reports SynthID and Content Credentials signals for OpenAI-made images.
Updated
How it works
When the extension sees an image, it sends the image URL to our API. The API fetches the file (up to 8 MB), reads its metadata, and sends the file to OpenAI's public content provenance endpoint. OpenAI answers with whether it finds its SynthID watermark and a valid OpenAI Content Credentials manifest. A positive answer is a provenance signal and pushes the score to at least 0.9; the card shows it under Image watermark with the badge Strong.
What OpenAI receives
The image file only. No account data, no extension token, no page hostname, no information about where the image appeared. OpenAI's handling of the file is governed by its API data usage policy. If you would rather no image reached a verifier, on-device only mode stops all of it.
Caching and budgets
Results are cached by content hash for 30 days, so a widely shared image is checked once. The check runs under a daily budget; when the budget is spent the row says the check was skipped rather than failing silently.
What it does not cover
Images made by Google, Midjourney, Stable Diffusion, Flux and others carry no OpenAI signal, so a negative answer here says nothing about them. Those are covered by Content Credentials, EXIF and IPTC metadata and the generator classifier.