OpenAI To Tag And Watermark Its Images, Launches Verification Tool Preview

OpenAI says it will start embedding standard provenance tags and an invisible watermark on pictures it creates, and will offer a public checker that can read both signals.

OpenAI announced Tuesday that it will begin marking images produced by its tools with a metadata tag based on the C2PA standard and add an unseen digital watermark developed with Google called SynthID. The company also showed a preview of a free verification service that can detect those markers, aimed at helping everyday users determine whether an image came from OpenAI’s systems. For now, the protections cover only images made by the company’s products.

C2PA, a nonprofit-backed protocol set up in 2021 to track content origins, attaches a clear identification field to a file’s metadata indicating it was machine-created. Because that tag sits where users and software can read and edit it, it can be altered or stripped out, making it most reliable among parties that already trust one another. Adoption of the standard has been spotty across the industry, limiting its reach.

SynthID is intended to be a tougher-to-remove layer. Built by Google and pledged to by OpenAI, the marker is embedded invisibly into the pixels so it survives common edits like screenshots, scaling or basic retouching. The two approaches are meant to compensate for each other’s shortcomings: one offers readable context, the other gives a resilient signal when files are transformed.

The measures won’t touch images churned out by smaller or dubious generators, and that gap matters. “This only fixes part of the problem,” said a digital forensics researcher who reviewed the announcement. “Unless other makers sign on, manipulated or synthetic visuals from other sources will still circulate without these safeguards.” An OpenAI representative told reporters the company hopes to encourage wider uptake and eventually add support for third-party images in its verification tool.

For consumers and newsrooms, the immediate benefit is a simple way to check OpenAI-origin images, but broad protection will depend on market-wide standards and enforcement. OpenAI’s step could nudge rivals and platforms to follow, yet experts warn that coordinated industry action, user education and stronger detection tools will be needed before digitally created imagery becomes consistently reliable.