OpenAI has announced an expansion of its content provenance efforts to help people identify and trust AI-generated media. Building on a multi-layered approach that combines open standards and watermarking, OpenAI updated its systems to include text provenance in the EU to comply with the EU AI Act. Additionally, supported audio generated through tools like ChatGPT and the OpenAI API now includes SynthID watermarking. OpenAI's public verification tool allows users to verify both images and audio files for provenance signals, and new API access enables developers and organizations to incorporate these checks into their workflows. OpenAI relies on a combination of C2PA conformance for metadata and Google DeepMind's SynthID for durable, invisible watermarking that survives modifications like screenshots or format changes. While the public verification tool can detect C2PA metadata and SynthID watermarks, it does not make definitive conclusions if no signals are detected, acknowledging that provenance data can sometimes be stripped. OpenAI has been involved in provenance standards since 2024, starting with DALL·E 3, and aims to support broader cross-industry verification efforts over time.
Advancing content provenance for a safer, more transparent AI ecosystem
OpenAI is expanding its content provenance initiatives by integrating C2PA standards, Google DeepMind's SynthID watermarking, and public verification tools. The updates span image and audio files generated through tools like ChatGPT and the OpenAI API. These measures aim to help users identify AI-generated media and comply with regulations like the EU AI Act.
Why it matters
Combining C2PA metadata and SynthID watermarking addresses the limitation that metadata can be stripped or lost during file transformations. This multi-layered approach helps establish a more durable and interoperable provenance ecosystem for AI-generated media across the industry.