OpenAI has announced two complementary measures intended to mark images produced by its artificial intelligence tools. The first uses the C2PA open standard: a metadata marker that records explicitly that a given image was generated by AI.
The second measure is SynthID, an invisible watermark implemented in collaboration with Google. This approach is designed to be harder to detect and remove than visible metadata tags, making it more difficult for malicious actors to strip or obscure the provenance information.
Preview verification tool
The company also released a preview version of a public verification tool that checks for both C2PA metadata and the SynthID watermark. At present, this verification tool can only recognize these signals in images produced with OpenAI’s systems.
Limitations and next steps
As TechCrunch points out, these new protections currently apply only to images generated with OpenAI’s tools; they do not immediately address images created by lesser-known or alternative AI solutions. OpenAI says it plans to expand the verification tool’s capabilities over time to cover images from additional sources.
Why this matters
AI-generated images are proliferating across the internet, and while some are easy to spot, others are so realistic that distinguishing them from genuine photographs with the naked eye is nearly impossible. OpenAI’s combination of C2PA metadata and the Google-backed SynthID watermark aims to make it more difficult to conceal the origin of such images, though it is not a complete solution until broader industry adoption occurs.
