AI Safety & Trust and Safety
Synthetic Media Policy
Purpose
Governs generation and disclosure of AI-generated synthetic media.
Scope
Image, audio, and video generation features.
Definitions
Synthetic Media.
Policy Statements
Media depicting real, identifiable individuals requires consent or contextual safeguards. C2PA-compatible provenance metadata is embedded where feasible (see Copyright & IP Policy, Doc. 5). Political deepfakes, non-consensual intimate imagery, and deceptive impersonation of public figures are prohibited.
Multimodal Risk Cross-Reference Detection and disclosure controls for cross-modal manipulation of synthetic media (e.g., steganographic instructions embedded in generated images/audio/video, or adversarial audiovisual content designed to bypass provenance checks) are governed jointly with the new Multimodal Risk Policy Doc. 38, which this policy now explicitly cross-references to close the previously one-directional gap between the two documents.
Responsibilities
Studio product team owns provenance implementation; Trust & Safety enforces restrictions.
Compliance Requirements
Provenance standards reviewed against evolving C2PA norms.
Enforcement
Removal and account suspension for violations.
Reporting Process
Exceptions Process
Creative/satirical exceptions require Trust & Safety review.
Review Process
Quarterly.
Contact Information
Related Policies
This policy references
Referenced by
None