AI-generated content disclosure
Overview
AI-generated content disclosure (also AI transparency labeling) refers to practices and requirements for indicating that content — text, images, audio, video — was produced wholly or substantially with the assistance of an AI system. Disclosure operates at the production layer (the creator declares) rather than through technical detection.
Disclosure practices range from voluntary editorial conventions (news organizations labeling AI-assisted drafts) to emerging legal mandates (EU AI Act requirements for certain AI-generated outputs). Standards for what constitutes sufficient disclosure remain unsettled: there is no consensus on whether disclosure should appear in metadata, as an inline label, in platform UI, or only when the AI's contribution is substantial.
Scope of disclosure
| Scope question | Contested? | Notes |
|---|---|---|
| AI-written vs AI-assisted | Yes | Minor grammar fix vs. majority-generated draft |
| Text vs. synthetic media | Partially | Image/video disclosure more standardized (C2PA); text less so |
| Platform vs. creator obligation | Yes | Some regulations target platforms; others target creators |
| Retroactive disclosure | Mostly no | Required at time of publication, not retroactively |
Regulatory landscape (as of 2025)
- EU AI Act (2024): Requires certain AI-generated or manipulated content (synthetic media, deepfakes in non-artistic contexts) to be marked. General-purpose AI systems must comply with transparency requirements.
- FTC (US): Existing guidance on deceptive advertising extends to undisclosed AI-generated endorsements.
- Platform policies: Meta, YouTube, TikTok, and others have policies requiring disclosure of synthetic media, particularly for political content.
- No binding international standard for AI-generated text disclosure exists as of this writing.
Distinction from AI content detection
| Dimension | AI content detection | AI-generated content disclosure |
|---|---|---|
| Mechanism | Technical inference from the text | Creator's own declaration |
| Reliability | Probabilistic, error-prone | Dependent on creator compliance |
| Regulatory driver | Limited | Increasing regulatory mandates |
Disclosure and detection are complementary: disclosure relies on honesty; detection is attempted where honesty cannot be assumed.
