OpenAI adds new labels to help identify AI made images

In short: OpenAI says it will make it easier to check whether an image was created by OpenAI’s own AI image tools.

What happened

OpenAI announced two steps meant to help people spot images made with its models. The goal is to make AI created images easier to identify, especially as fake or misleading images spread online.

First, OpenAI is joining C2PA, an open standard for content labels. C2PA works by adding information to a file’s metadata, which is extra data stored with a photo (like a label inside the packaging). That metadata can include a clear note that an image was generated by AI.

Second, OpenAI will add Google’s SynthID to its products. SynthID is an invisible watermark, which is a hidden marker inside the image (like ink you cannot see, but a scanner can). OpenAI says this marker is designed to stay even if someone edits the image, takes a screenshot, or resizes it.

OpenAI also previewed a public verification tool that checks for both signals. At first, the checker will only work for images generated by OpenAI products, not images made by other AI tools.

Why it matters

Many people now see images online and cannot tell if they are real, edited, or fully computer made. Adding both a visible style label in metadata and a harder to remove hidden watermark could make it easier for journalists, platforms, and everyday users to double check what they are looking at. It also sets expectations that AI companies should help track the content their tools create.

Source: TechCrunch AI