SynthID and C2PA are being tested to see if they can effectively combat deepfakes. These technologies, which tag image, video, and audio files with origin information, are getting a major expansion, offering a chance to counter unlabeled AI content deceiving people online. At its I/O conference, Google announced that checking for SynthID markers will become available on Chrome and Search, a move that puts AI verification tools in front of many users. This will simplify the process of checking images for SynthID markers, which currently requires uploading to the Gemini app.
Google’s new verification interfaces will also identify C2PA information, metadata embedded during creation that indicates how and whether AI was used to make or alter the content. Users can now check suspicious images without switching between the Gemini app and C2PA portals, as both label types can be examined from one interface.
Both Google and the Content Authenticity Initiative, which promotes the C2PA standard, emphasize the need for widespread adoption of these systems. AI models need to embed this data, and platforms where AI-generated content is shared need to display this information clearly. Having verification tools built into browsers serves as a workaround for platforms not offering AI metadata.
OpenAI is participating by embedding SynthID into images produced by ChatGPT, Codex, and the OpenAI API. Although OpenAI already integrates C2PA metadata into content, this is often removed when posted on other platforms. Despite OpenAI’s commitment to C2PA, it acknowledges the challenges of metadata, which can be stripped accidentally or intentionally.
Though C2PA is touted as the standard for content authenticity, it hasn’t conclusively proven reliable in the wild. SynthID, with its resistance to removal, seems more promising, as there have been instances where it helped debunk deepfakes online. Both technologies should complement each other rather than compete, with Google having the chance to show its system’s reliability.
Meta plans to use C2PA metadata to tag Instagram images captured by cameras, potentially indicating that an image was “captured on Pixel 10,” assisting in distinguishing real photos from AI-generated ones.
Instagram is already checking images for C2PA information. However, previous attempts to label AI-generated content on the platform mistakenly tagged genuine photographs, causing controversy.
Despite the opportunity presented by this collaboration, Google’s dual role as both the creator of AI technologies and enforcer of transparency raises concerns. Success of SynthID in countering deepfakes could offset these concerns, but expectations remain cautious due to the complexity of the issue.
Ultimately, SynthID and C2PA can only identify watermarks if they are present, and it’s unlikely that open-source models responsible for harmful deepfakes will adopt these systems. While provenance isn’t a perfect fix, Google and C2PA now have the chance to show it’s worthwhile.