Page 20 - Detecting deepfakes and generative AI: Report on standards for AI watermarking and multimedia authenticity workshop

Detecting deepfakes and generative AI: Report on standards for AI
watermarking and multimedia authenticity workshop

5 Session 3: AI watermarking, multimedia authenticity, and

provenance

The main objective of this session was to focus on multimedia provenance and provide examples
of industry initiatives and areas where standards are needed. This session examined industry-
led initiatives such as C2PA's Content Credentials specification. The session also considered
how a combination of AI watermarking to identify AI-generated multimedia and Content
Credentials tied to a real-world identity via cryptography could help in protecting copyright.

Moderator: Emma Brown, DeepMedia

Speakers

• Leonard Rosenthol, Chief Architect, Adobe.
• Andrew Jenks, Director of Media Provenance, Microsoft, and Chair of C2PA.
• Hu Zhengkun, Director of AI Ethics and Governance Research, SenseTime.
• Jon Geater, Chief Product and Technology Officer, DataTrails, and Co-Chair of the IETF
SCITT Working Group.
• Touradj Ebrahimi, Professor at EPFL, Executive Chairman of RayShaper SA, and Chair of
JPEG.

Provenance – the record of AI-generated content's origin and history – is crucial for verifying
authenticity, protecting creators' rights, and enhancing content quality.

Authentication or provenance verification – the process of assessing content's accuracy and
consistency – can help combat disinformation and ensure the credibility of multimedia content.
They can be used to enhance transparency and accountability in various industries by providing
a secure and transparent way to track the provenance and ownership of digital assets.

Leonard Rosenthol, Chief Architect at Adobe, provided an overview of C2PA's Content
Credentials specification. Andrew Jenks, Director of Media Provenance at Microsoft and
Chair of C2PA, explained that the specification is an example of affirmative authenticity which
represents a cryptographically secure method of marking content.

12