OpenAI joins the Board of Directors of the Content Provenance and Authenticity Alliance (C2PA) and is integrating open metadata standards into AI design standards to ensure transparency in content creation.
The C2PA standard allows digital content to be authenticated with metadata that proves its origin, whether created by AI, modified using AI tools, or captured in a traditional way. OpenAI has started adding C2PA metadata to output images from the latest DALL-E 3 model in ChatGPT and the OpenAI API. Metadata will be integrated into OpenAI’s next video production model, Sora, once it becomes more available.
“In the absence of this information, people can still create (or delete) fake content, but they cannot easily edit or modify the content, making it important for building trust,” OpenAI explained.
The move comes amid growing concerns that AI content could mislead voters ahead of major elections in the US, UK and other countries this year. Analysis of AI-generated news can help against deepfakes and other elements used to check fake campaigns.
While these measures are useful, OpenAI recognizes that achieving real content in practice requires coordination between platforms, creators, and content systems to store metadata for the end customer.
While these measures are useful, OpenAI recognizes that achieving real content in practice requires coordination between platforms, creators, and content systems to store metadata for the end customer.
OpenAI has opened access to DALL-E 3 image sensing applications through its Open Access Program. The tool estimates the probability that the image is derived from one of OpenAI’s models.
Internal testing demonstrated the accuracy of distinguishing non-AI images from DALL-E 3 detection; Approximately 98% of DALL-E images were identified correctly, and less than 0.5% of the image was misregistered by the AI. However, the classifier has many differences between images generated by DALL-E and other generative AI models.
OpenAI has also added watermarking to its custom speech engine, which is currently in limited preview.
The company believes that using increasingly proven standards will make metadata along with content throughout its lifecycle and make a “significant difference in enforcing the accuracy of content.”
OpenAI and Microsoft announced a $2 million Social Resilience Fund to promote artificial intelligence and understanding through AARP, International IDEA, and the Artificial Intelligence Partnership.
“While the solutions described above provide us with protection tools, the practical implementation of reality requires collaboration,” OpenAI said.
Our provenance efforts are just one part of broader industry efforts; Many of our research partners and prolific AI companies are also accelerating research in this area. We appreciate these efforts; It is important for businesses to collaborate and share insights to improve our understanding and maintain online transparency.
For the latest tech news, follow Gadgetsfocus on Facebook, and TikTok. Also, subscribe to our YouTube channel.