OpenAI is rolling out a broader set of tools meant to make it easier to identify whether an image was created with its systems. The company said it is strengthening its content provenance approach with new standards support, watermarking through a Google partnership, and a preview of a public verification tool.
The update comes as AI-generated media becomes more common across chatbots, design tools and creative workflows. OpenAI said provenance signals can give people more context about how an image was made, edited or shared, and help them judge online content with greater confidence.
At the center of the effort is a push to make those signals easier for other platforms and tools to read. OpenAI said it has become a conforming generator under C2PA, a cross-industry standard for content provenance. The standard uses metadata and cryptographic signatures to help information about a file travel with the file itself. In practice, that can help platforms preserve details about origin and creation history as content moves across services.
OpenAI has been working with provenance standards since 2024, when it began adding Content Credentials to images made with DALL·E 3 and later extended the practice to ImageGen and Sora. By aligning more closely with C2PA, the company is trying to make its provenance information more durable as images are uploaded, downloaded and reposted.
The company is also adding another layer of detection through Google DeepMind’s SynthID, starting with images generated in ChatGPT, Codex and the OpenAI API. SynthID places an invisible watermark into images, which OpenAI says can survive more transformations than metadata alone. That matters because file compression, resizing, screenshots and format changes can strip or damage metadata.
OpenAI described the move as a multi-layered system, with metadata and watermarking intended to reinforce each other. Metadata can provide richer context, while watermarking can remain detectable even when file information is lost. The company said it has already used visible watermarks in Sora and audio watermarking in Voice Engine as part of its broader safety and transparency work.
Alongside those changes, OpenAI is previewing a public verification tool that lets people upload an image and check whether it appears to have been generated by its products. The tool looks for both Content Credentials and SynthID signals. If it finds evidence, it can indicate that the image was made with ChatGPT, the OpenAI API or Codex.
OpenAI said the verifier is intentionally cautious. If no metadata or watermark is found, it will not declare that an image was or was not created with OpenAI tools. The company noted that provenance signals can be removed, so absence of evidence is not treated as proof either way.
For now, the verification tool only works with OpenAI-generated content. The company said it hopes to expand support over time and work with industry partners on broader cross-platform verification.
OpenAI framed the changes as part of a larger effort to build a more trustworthy ecosystem for online media. It said no single approach is enough on its own, and that durable provenance will likely depend on a mix of shared standards, watermarking and public verification tools.