AI Safety & Trust and Safety
Content Policy
Purpose
Defines standards for content generated, uploaded, or shared through Services.
Scope
Text, image, audio, video, and code content.
Definitions
Restricted Content.
Policy Statements
Prohibits CSAM, non-consensual intimate imagery, credible threats of violence, and hate content. Restricted content permitted with age verification and safeguards. Synthetic media depicting real individuals follows the Synthetic Media Policy.
Multimodal Risk Coverage Content moderation classifiers now explicitly cover cross-modal jailbreak vectors: image-embedded text instructions, audio-based prompt injection, and video-frame manipulation designed to bypass text-based filters. See new Multimodal Risk Policy Doc. 38 for detailed controls.
Responsibilities
Trust & Safety and Content Moderation own enforcement; Engineering maintains multimodal classifier tooling.
Compliance Requirements
Moderation systems audited quarterly, including multimodal-specific bias and accuracy testing.
Enforcement
Removal; repeat violations lead to account action.
Reporting Process
Exceptions Process
Research/journalistic exceptions require Trust & Safety approval.
Review Process
Quarterly.
Contact Information
Related Policies
This policy references
Referenced by
None