The misuse of generative AI products for spreading disinformation, OpenAI has announced the integration of C2PA metadata into images created by DALL-E 3. C2PA, an open technical standard, enriches images with origin details, edit history, and other pertinent information. OpenAI states that images generated through DALL-E 3, directly or via ChatGPT or its API, will now carry C2PA data, with implementation for mobile users scheduled by February 12. The inclusion of C2PA metadata offers benefits in image verification, enabling uploads to platforms like Content Credentials Verify to authenticate image provenance. Social media platforms, including Meta, could leverage this data to identify and label AI-generated images. OpenAI acknowledges that while C2PA is not foolproof against provenance issues, it serves as a step towards bolstering the trustworthiness of digital information.

The addition of C2PA data does come with trade-offs, as it increases file sizes. OpenAI provides estimates, citing a 3% increase for PNG through API, a 5% increase for WebP through API, and a notable 32% increase for WebP through ChatGPT. Despite these efforts, OpenAI recognizes the ease with which C2PA metadata can be removed, even unintentionally, through actions such as taking a screenshot or converting file formats. While OpenAI hopes that integrating C2PA into DALL-E images will encourage users to recognize authenticity signals, critics argue its susceptibility to removal limits its effectiveness against intentional misuse. Comparisons are drawn to Google's SynthID digital watermark, perceived as a more secure solution, although OpenAI, being a competitor, is unlikely to adopt it. As the battle against disinformation rages on, the efficacy of metadata solutions remains a topic of ongoing debate.