AI Content Watermarking and Labeling Compliance

Key steps

• Map the operating jurisdictions to their AI content labeling requirements. Current key requirements:

  • EU AI Act: AI-generated or manipulated content (deepfakes, synthetic audio/video) must be labeled as artificially generated. General-purpose AI (GPAI) model providers must publish summaries of training data.
  • China AIGC (AI-generated content) Measures: AI-generated content must carry visible labels indicating it was produced by AI. Platforms must implement labeling at the point of generation and in metadata (identifying information stored inside the file).
  • China Deep Synthesis Regulations: synthetic media (voice, video, text) must carry explicit disclosure. Service providers must implement watermarking (marks embedded in content that identify it as AI-made).
  • US: No current federal mandate, but Federal Trade Commission (FTC) guidance treats misleading AI content as unfair or deceptive. State laws vary.

• Define the labeling approach for each content type: text outputs, images, video, audio, synthetic voice.

• Implement technical controls: label templates, API flags (settings in the systems that generate content) that trigger labeling, metadata watermarking for jurisdictions that require it.

• Train content production teams on when labeling is required and what format to use.

• Test labeling controls quarterly and review audit logs for unlabeled AI outputs.

Common gaps

• Applying labels to customer-facing content but not to internal reports, board presentations, or regulatory filings that use AI-generated content.
• Treating watermarking as optional when some jurisdictions make it mandatory.
• Missing jurisdiction-specific labeling format requirements (some regulations specify where the label must appear, not just that one must exist).