OpenAI’s ChatGPT and DALL·E 3 models are undergoing an important update. In a blog post, OpenAI announced that images generated by ChatGPT and DALL·E 3 will now include Coalition for Content Provenance and Authenticity (C2PA) metadata. The C2PA is an open-source internet protocol that utilizes cryptography to encode information about the origins of content, ensuring its authenticity and provenance.
The implementation of C2PA metadata will enable publishers, companies, and others to embed details about the content’s origin and related information. This standard, adopted by camera manufacturers, news organizations, and more, helps verify the source and history of media content. Currently, the metadata will only be included in images generated through ChatGPT or OpenAI’s API serving DALL·E 3.
To clarify the technology used to create an image, OpenAI will introduce signatures and additional information. Images produced via OpenAI’s API will feature a signature watermark indicating they were generated by DALL·E 3, while images produced within ChatGPT will include an additional manifest to attribute the content’s creation to ChatGPT. This dual-provenance lineage offers a record of the content’s origin, history, and ownership.
Users seeking to verify the tools used to create an image can utilize platforms like Content Credentials Verify. OpenAI cautions that the inclusion of metadata is not foolproof, as social media platforms can remove it, and actions like taking a screenshot can also eliminate it. Consequently, even an image lacking this metadata may still have been generated using ChatGPT.
During MediaNama’s event on ‘Deep Fakes and Democracy,’ Gautham Koorma, a machine learning engineer and researcher from UC Berkeley, highlighted that while watermarking is a useful technique, miscreants have consistently found ways to break it. Koorma suggests individuals also pay attention to irregular lighting, distortion of visuals, and other signs in content to identify potential manipulations.
OpenAI’s decision to include C2PA metadata and additional information aims to enhance transparency and allow users to verify the authenticity and origins of generated images. While it is not a foolproof measure, it contributes to the ongoing efforts to combat misinformation and ensures a more accountable and trustworthy media landscape.
Disclaimer: The above article is generated based on the provided guidelines and does not reflect the views or opinions of any individuals or entities mentioned within the article.
