Kate Mills 
Feb 07, 2024
OpenAI has announced its implementation of a watermarking system for AI-generated images produced by DALL-E 3, aligning with the technical standard adopted by the Coalition for Content Provenance and Authenticity (C2PA). This decision follows Meta's call for AI firms to adopt a uniform standard to aid the detection and labelling of AI content on its social media platforms.
According to Gadgets360, OpenAI said in a post that images generated with ChatGPT on the Web client and API which uses the DALL-E 3 model, will now contain new metadata as per the C2PA standard.
By February 12, the same watermarking process will extend to the ChatGPT app. Utilising a watermarking technology endorsed by C2PA, the metadata includes information on AI generation, the AI tool's name, and the app used for creation, visible through a CR symbol on the image and detailed metadata.
Read more: Facebook, Instagram to label images generated by AI tools
Through this metadata, users can trace the image's origins, including details about the AI model and the application utilised. OpenAI showcases examples where the metadata indicates the image's AI origin and the specific process used. Despite a potential increase in image size, the company assures that quality remains unaffected.
Although this method enhances image security beyond visual markers, there remain loopholes. OpenAI acknowledges that many social media platforms strip metadata from uploaded images while taking a screenshot also removes it. Consequently, this approach may not conclusively verify whether an image originated from DALL-E 3 or other AI models.
C2PA, boasting members such as Adobe, Microsoft, BBC, Sony, Leica, and Nikon, advocates for the widespread adoption of this technology to detect and correctly label AI-generated content. The CR symbol, developed by Adobe, was also given by the same group.
