OpenAI is making it easier to check if an image was generated by its models

With AI-powered image generators widely available online and more sophisticated than ever before, it’s never been harder to tell if an image is authentic or not. But on Tuesday, OpenAI announced two new measures to help combat the problem.

The company has committed to an open standard called C2PA, which adds a clear indication in the metadata that the image was generated by artificial intelligence. OpenAI is also partnering with Google to include an invisible watermark called SynthID, which will be difficult to detect, but also difficult to erase if bad actors try to cover their tracks.

The new protections only apply to images generated by OpenAI products, so they won’t affect the flow of images coming from less popular AI tools; They can help ensure that OpenAI is not part of the problem.

OpenAI is also previewing a public verification tool that will check both signals, allowing users to easily test whether an image was created with AI. Initially, the tool will only extend to images generated by OpenAI products; The company hopes to expand it to other tools over time.

Founded in 2021, the Alliance for Content Procurement and Authenticity (C2PA) is a non-profit organization dedicated to mitigating the harmful effects of AI imagery on public discourse. The C2PA standard has been adopted by a range of Google products, but its adoption remains inconsistent across the industry. Since the C2PA signal is clearly accessible in the metadata of each file, it can be manipulated, and is very useful among trusted users.

SynthID is a newer effort designed to be a more robust intervention measure. Developed by Google, the SynthID watermark is designed to persist even when bad actors try to remove it, either through screenshots, resizing, or digital manipulation.

The two systems are intended to complement each other, as each addresses the other’s weaknesses.

“Watermarking can be more sustainable through transitions such as screenshots, while metadata can provide more information than a watermark alone,” OpenAI noted in its announcement. “Together they make the source more flexible than any layer could be alone.”