OpenAI Adds Metadata Standard and Watermark to AI Images

OpenAI announced two measures on Tuesday to help identify images created by its AI tools. The company said it will join the Coalition for Content Provenance and Authenticity standard and add Google's SynthID watermark to its generated images. The C2PA standard inserts metadata into image files that indicates they were produced by AI.

OpenAI stated the metadata signal will be accessible to users who check file properties. OpenAI is also incorporating SynthID, an invisible watermark developed by Google. The company said the watermark is designed to remain detectable after common edits such as resizing or screenshots.

OpenAI previewed a public verification tool that checks for both the C2PA metadata and SynthID watermark. The tool will initially cover only images made with OpenAI products. The company said it plans to expand the tool to images from other AI generators over time.

Provenance and Authenticity was founded in 2021 as a non-profit focused on content authenticity. OpenAI noted that the two systems address different vulnerabilities, with metadata providing more details and the watermark offering greater durability.

"Watermarking can be more durable through transformations like screenshots, while metadata can provide more information than a watermark alone," OpenAI stated in its announcement. " The new measures apply only to images generated by OpenAI products and do not cover content from other AI tools.