OpenAI is joining the Coalition for Content Provenance and Authenticity (C2PA) steering committee and will integrate the open standard’s metadata into its generative AI models to increase transparency around generated content.
The C2PA standard certifies digital content with metadata proving its origins, whether created entirely by AI, edited with AI tools, or captured traditionally. OpenAI has already begun incorporating C2PA metadata into images from its latest DALL-E 3 model in ChatGPT and the OpenAI API. This metadata will also be integrated into OpenAI’s upcoming video generation model, Sora, upon its broader release.
“People can still create deceptive content without this information (or can remove it), but they cannot easily fake or alter this information, making it an important resource to build trust,” OpenAI explained.
This move comes amid rising concerns about AI-generated content potentially misleading voters ahead of major elections in the US, UK, and other countries this year. Authenticating AI-created media could help combat deepfakes and other manipulated content aimed at disinformation campaigns.
While technical measures are essential, OpenAI acknowledges that enabling content authenticity in practice requires collective action from platforms, creators, and content handlers to retain metadata for end consumers.
In addition to C2PA integration, OpenAI is developing new provenance methods such as tamper-resistant watermarking for audio and image detection classifiers to identify AI-generated visuals.
OpenAI has opened applications for access to its DALL-E 3 image detection classifier through its Researcher Access Program. This tool predicts the likelihood that an image originated from one of OpenAI’s models.
“Our goal is to enable independent research that assesses the classifier’s effectiveness, analyses its real-world application, surfaces relevant considerations for such use, and explores the characteristics of AI-generated content,” the company said.
Internal testing shows high accuracy in distinguishing non-AI images from DALL-E 3 visuals, with around 98% of DALL-E images correctly identified and less than 0.5% of non-AI images incorrectly flagged. However, the classifier has more difficulty differentiating between images produced by DALL-E and other generative AI models.
OpenAI has also incorporated watermarking into its Voice Engine custom voice model, currently in limited preview.
The company believes that increased adoption of provenance standards will ensure metadata accompanies content throughout its lifecycle, filling “a crucial gap in digital content authenticity practices.”
OpenAI is partnering with Microsoft to launch a $2 million societal resilience fund to support AI education and understanding, including through AARP, International IDEA, and the Partnership on AI.
“While technical solutions like the above give us active tools for our defenses, effectively enabling content authenticity in practice will require collective action,” OpenAI states.
“Our efforts around provenance are just one part of a broader industry effort—many of our peer research labs and generative AI companies are also advancing research in this area. We commend these endeavors—the industry must collaborate and share insights to enhance our understanding and continue to promote transparency online.”
Leave a comment