OpenAI steps up fight against deepfakes with watermarking and verification tools

The new system enables users to upload images and check for embedded AI-origin signals, marking OpenAI’s push toward greater content traceability as concerns rise over the rapid spread of highly realistic synthetic visuals online.

OpenAI has introduced new safeguards designed to curb the misuse of AI-generated images and improve the reliability of digital content circulating online.

Announced on May 19, the update includes a collaboration with Google and introduces a public verification system aimed at helping users determine whether an image has been created using AI tools.

Watermarking and detection system

As part of the initiative, OpenAI will integrate Google’s SynthID technology, which embeds invisible watermarks into AI-generated images without altering their appearance. The company is also adopting C2PA metadata standards, which attach secure digital labels to files to indicate AI involvement.

These combined methods are intended to strengthen traceability and make it easier to identify synthetic content. Alongside this, OpenAI has launched a verification tool that allows users to upload images and check whether they were generated using its platforms.

Currently, the system supports images created through ChatGPT, Codex, and OpenAI’s API services. However, OpenAI has indicated that broader support across additional AI systems is planned in the coming phases.

“At launch, the tool is limited to content generated by OpenAI. In the upcoming months, we aim to support cross-industry efforts to make verification possible across platforms,” the company stated in a blog post.

Addressing rising synthetic media risks

The initiative comes amid growing concern over the increasing realism of AI-generated visuals, which are becoming harder to distinguish from authentic photographs.

While the system is expected to improve transparency, OpenAI acknowledges that independent or lesser-regulated AI tools may still pose challenges in combating misinformation.

Users are advised to upload original, unedited files for accurate verification, as heavily modified images may reduce detection effectiveness. The system primarily scans for SynthID watermarks and C2PA metadata to determine whether AI tools were involved in image creation.