OpenAI has introduced a new public verification tool designed to help people check whether an image was likely generated using OpenAI tools such as ChatGPT, the OpenAI API or Codex. The tool is part of a broader push by the company to make AI-generated media easier to identify at a time when synthetic images are moving quickly across social platforms, news feeds and private chats.

The verification tool does not look at an image and decide whether it is “real” or “fake” in the general sense. Instead, it searches for provenance signals that OpenAI attaches to supported AI-generated images. These signals include C2PA Content Credentials and SynthID watermarks, two different methods meant to help trace where a piece of media came from.

For everyday users, the idea is simple. Upload an image, wait for the tool to scan it, then review whether OpenAI’s system detects a supported signal. For journalists, researchers, platforms and ordinary internet users, the tool could become one more check before sharing or reporting on an image that appears suspicious, surprising or politically sensitive.

What OpenAI’s Image Verification Tool Does

OpenAI’s new tool is currently available as a research preview. According to OpenAI, users can upload an image to check whether it contains signals associated with images generated by OpenAI products. The supported file formats listed on the tool page are PNG, JPG and WEBP.

The tool checks for two main signals. The first is C2PA metadata, also known as Content Credentials. This is a provenance standard that can attach information to a media file, including information about where the content came from and how it was created or edited. OpenAI says C2PA uses metadata and cryptographic signatures to help that information travel with digital content.

The second signal is SynthID, an invisible watermarking system from Google DeepMind. OpenAI says SynthID embeds a signal directly into generated media, making it different from ordinary metadata, which can be removed when an image is uploaded, downloaded, converted or screenshotted.

Together, the two systems give OpenAI a layered approach. C2PA can carry detailed information about the origin of a file. SynthID can remain useful when metadata is lost or stripped. OpenAI says images generated with ChatGPT, Codex and its API include both C2PA metadata and SynthID watermarks.

How To Use OpenAI’s Image Verification Tool

To use the tool, go to OpenAI’s image verification page and upload a single image. OpenAI says users can drag and drop a file or click to upload it. The page currently supports PNG, JPG and WEBP files.

After the upload, the tool scans the image for supported provenance signals. It may report that it found C2PA metadata, a SynthID watermark or no supported signal. OpenAI recommends uploading only one image at a time. If the image is a screenshot, users should crop it closely around the image they want to check and avoid uploading files that contain several pictures at once.

A detected signal means the image likely originated from OpenAI tools. OpenAI says detected signals are reliable and that false positives are rare. However, the company is careful about what the result means. A positive match does not prove that the image is accurate, unedited, legally owned or being shared in the right context.

A “no signal found” result should also be read carefully. It does not prove that an image is human-made. It only means the tool did not detect OpenAI’s supported provenance signals. OpenAI says an image could still have been generated by its tools if metadata was stripped, a watermark was degraded, the image came from an unsupported or older source, or it was created before these signals were available. It could also have been made by another company’s AI model, which the current OpenAI tool does not detect.

OpenAI also says uploaded images are processed to check for supported provenance signals and are not stored unless legally required. The company says uploads to the verification tool are not used to train its models.

Why This Matters

The internet has entered a period where images can no longer be judged by appearance alone. AI tools can create convincing scenes, edit real photos and generate content that looks like a screenshot, a news photo or a personal memory. That has made provenance, the record of where a piece of media came from, a key part of online trust.

OpenAI’s move is not just about giving users a button to press. It is also about making AI-generated media easier for other systems to read and preserve. The company says it has become a C2PA Conforming Generator Product, which means platforms should have a more trusted way to read and pass along the provenance information attached to OpenAI-generated content.

This matters because metadata can be fragile. A social media upload, screenshot or file conversion can remove important information from an image. That is one reason OpenAI is pairing C2PA with SynthID watermarking, which is designed to provide a more durable signal within the image itself.

For newsrooms, this kind of tool can be useful during breaking news. A viral image may appear before reliable reporting catches up. A provenance check will not replace verification, but it can help editors decide whether a photo deserves extra caution. For ordinary users, it can be a quick way to check whether an image that looks suspicious carries OpenAI’s own generation signals.

What The Tool Cannot Do

The tool has important limits. At launch, it is limited to content generated by OpenAI. It is not a universal AI-image detector. If an image was created by another AI company’s model, OpenAI’s tool may not identify it.

It also does not judge truthfulness. A detected signal only says that the image likely came from OpenAI tools. It does not say whether the image has been edited later, whether the caption is honest or whether the scene is being used to mislead people.

That distinction is important. A real image can be used deceptively with a false caption. An AI-generated image can be used transparently for satire, illustration or education. Provenance helps answer the question of origin, but it does not answer every question about intent, context or accuracy.

OpenAI says no detection method is foolproof. If no watermark or metadata is detected, the tool will not make a definitive conclusion about whether the image was created using OpenAI tools. That cautious approach is necessary because provenance signals can be removed, damaged or absent from older content.

The Bigger Picture

OpenAI’s verification tool fits into a wider industry effort to label and trace synthetic media. In 2024, OpenAI said it was adding C2PA metadata to images created and edited by DALL·E 3 in ChatGPT and the OpenAI API. The company also opened access to an image detection classifier for selected testers, including research labs and research-oriented journalism nonprofits.

The new public tool brings part of that effort closer to ordinary users. Rather than relying only on researchers or platforms, OpenAI is giving the public a way to check whether an image contains signals tied to its own systems.

The company says it expects to support more types of content over time and wants to work with broader cross-industry efforts so verification can eventually work across platforms.

For now, the practical takeaway is straightforward. OpenAI’s verification tool is useful, but it should be treated as one signal in a larger verification process. It can help identify images that likely originated from OpenAI tools. It cannot prove that an image is true, complete, unedited or harmless. In the age of AI-generated media, that is still a meaningful step, but it is not the final word.