Per a report of OpenAI's announcement, images get an invisible watermark, and No Film School says the Sora watermark could be removed

According to a reproduction of OpenAI's own announcement, the company has introduced Google's SynthID watermark along with so-called C2PA content credentials on images from ChatGPT, Codex and the API, making it easier to tell whether an image was made with AI. The details come from intermediaries, and OpenAI's own announcement could not be read directly. The sources do not make clear whether SynthID also covers video.

SynthID is, according to AI Weekly, an invisible watermark that survives cropping and compression. C2PA is a kind of digital sender that stores information about where an image comes from. According to a reproduction of OpenAI's announcement, OpenAI is also registered as a C2PA-compliant generator product, so provenance data can be more easily preserved across platforms.

Tool for checking images

OpenAI has also shown a preview of a public tool where you can upload an image and see whether it contains Content Credentials or SynthID. At first, it only works for images from OpenAI itself.

If the tool finds no signal, it does not give a definitive answer, because provenance data can be removed in some cases. OpenAI has not explained how the verification handles heavy editing or recompression. According to AI Weekly, OpenAI plans to support industry-wide solutions so content can be verified across platforms in the coming months.

Sora watermark could be removed, according to No Film School

According to No Film School, all Sora videos have a visible watermark and C2PA metadata at launch. No Film School also repeats OpenAI's own claim that the company's internal reverse image and audio search tools can trace videos back to Sora with high accuracy.

No Film School also writes that the watermark on Sora 2 videos could be removed about five days after launch, while the app had passed one million downloads. That is the outlet's own description and not an independent test.

Author Jason Hellerman believes the solution was naively designed and that it creates a risk of fraud, fake news and deepfakes of public figures. That is his assessment. He writes: "We can't even trust them to make watermarks work, so why do we trust them to do anything else?"

Complaints about deepfakes

Sora launched on September 30, 2025. After complaints from actor Bryan Cranston and the families of Robin Williams, George Carlin and Martin Luther King Jr., OpenAI tightened protections against videos using celebrities' likeness and voice without consent, MacRumors reports.

Representatives of deceased historical figures can ask that the person's likeness not be used, and OpenAI has promised to respond quickly to complaints. According to MacRumors, a joint statement from October 20, 2025, including SAG-AFTRA and Bryan Cranston, says that artists, performers and individuals should have the right to decide whether and how they can be simulated with Sora.