OpenAI has introduced new tools to detect and mark AI-generated content.
These include an image detection classifier that identifies if photos were created by its DALL-E 3 generator, boasting about 98% accuracy, though it struggles with content from other AI models.
Additionally, a tamper-resistant watermark for audio and other content has been developed.
OpenAI is also enhancing content credentials with information about image ownership and creation.
These advancements, part of efforts with the Coalition of Content Provenance and Authority, aim to improve the traceability of AI-generated media. Feedback from users and tests by researchers will help refine these tools.