TLDR:
- OpenAI has announced that it will add “Made by AI” metadata to the output of its generative models, in order to disclose their origin.
- This move is aimed at alerting people to the fact that the content they are viewing was created by AI, helping to distinguish between real and fake images.
- The metadata is not directly visible to users and is cryptographically protected to prevent unauthorized changes. Applications that support this metadata will display a logo indicating the presence of Content Credentials information.
- However, the Content Credentials strategy has limitations. The metadata can be easily removed or cropped out, and it relies on apps and services to support the specification.
OpenAI has announced that it will add “Made by AI” metadata to the output of its generative models, in an effort to disclose their origin and help users distinguish between real and fake images. The Microsoft-backed AI lab will adopt the Content Credentials specification developed by the Coalition for Content Provenance and Authenticity (C2PA), which uses standard data formats to store details about who made the material and how. While the metadata is not visible to users and is protected from unauthorized changes, it can be easily stripped out or removed, and its effectiveness relies on apps and services supporting the specification. OpenAI acknowledges that the metadata is not a foolproof solution but believes that it can help increase the trustworthiness of digital information.
The introduction of “Made by AI” metadata comes as concerns grow over the spread of synthetic media and the potential for misinformation and disinformation campaigns. OpenAI’s decision to include this metadata in its generative models aims to address these concerns and provide users with information about the origin of the content they are viewing. By displaying a logo indicating the presence of Content Credentials information, supporting applications can make it clear to users that the content was generated by AI. However, the effectiveness of this strategy depends on app and service support, as well as user awareness and understanding of the metadata.