AI Disclosure
AI disclosure is the practice of marking content as AI-generated or AI-assisted so audiences, platforms and regulators can tell. It works on two layers: a human-readable label and a machine-readable mark embedded in the file. In 2026, two laws put firm dates on the machine-readable layer: Article 50 of the EU AI Act and California's AI Transparency Act, both starting August 2, 2026.
What are the two layers of AI disclosure?
AI disclosure has a human-readable layer that people see and a machine-readable layer that software detects. The table below shows how they differ.
| Layer | What it is | Examples | Who usually owes it |
|---|---|---|---|
| Human-readable | A visible label or note telling people AI was used | Caption, on-image label, stated note | The organization publishing the content |
| Machine-readable | A technical mark that software can read to tell the content was created or edited by AI | Metadata, invisible watermark, C2PA content credentials | The provider of the AI system that generates the content |
Regulation is moving fastest on the second layer, because labels alone are easy to remove and hard for platforms to verify at scale.
What does the EU AI Act require?
Article 50 of the EU AI Act sets transparency duties for AI systems and AI-generated content, and it applies from August 2, 2026. Per law firm summaries, providers must tell users when they are dealing with an AI system and mark AI-generated content so it can be detected as such. Deployers must flag sensitive uses, such as emotion recognition, and must label deepfakes and AI-generated text published on matters of public interest.
- The Digital Omnibus on AI (Regulation (EU) 2026/1744), in force since July 27, 2026, delayed most high-risk obligations but did not postpone Article 50.
- Generative systems already on the market before August 2, 2026 have until December 2, 2026 to meet the Article 50(2) marking duty. Systems launched after that date get no grace period.
- A later date, February 2, 2027, applies to watermark-detection interoperability under Article 50(2).
- The European Commission published Article 50 guidelines on July 20, 2026 and a voluntary code of practice on marking AI-generated content on June 10, 2026.
Some commentary also describes an editorial-control exemption for AI-assisted text that a human reviews and takes responsibility for. Confirm how it applies to a specific workflow before relying on it.
What does California's AI Transparency Act require?
California's AI Transparency Act (SB 942, as amended by AB 853) requires large generative AI providers to disclose when image, video or audio content is machine-generated. It covers providers whose systems have more than one million monthly visitors or users and are publicly accessible in California. Since August 2, 2026, covered providers must:
- Embed a latent (machine-readable) disclosure in AI-generated image, video and audio content.
- Offer users the option to add a visible, manifest disclosure.
- Make a free, public AI detection tool available.
Law firm summaries report a penalty of $5,000 per violation per day. A text-only chatbot is not covered for its text output, but the same product is covered once it generates images, video or speech. AB 853 also added later duties for large online platforms and hosting platforms (January 1, 2027) and for capture-device makers (January 1, 2028).
What are the key dates?
The key dates run from 2024 to 2028, and the table lists the ones most relevant to brands and AI providers.
| Date | Event |
|---|---|
| September 19, 2024 | California governor signs SB 942 |
| October 13, 2025 | AB 853 signed, moving California's start date from January 1, 2026 to August 2, 2026 |
| August 2, 2026 | EU Article 50 applies; California duties for covered providers become operative |
| December 2, 2026 | End of EU grace period for marking by generative systems already on the market |
| January 1, 2027 | California duties begin for large online platforms and hosting platforms |
| February 2, 2027 | EU watermark-detection interoperability duty |
| January 1, 2028 | California duties begin for capture-device manufacturers |
What should a brand do about AI disclosure?
A brand should treat AI disclosure as a content-governance task: know where AI is used, decide when to label it, and keep the machine-readable marks intact. Common steps include:
- List the images, video, audio and text that use AI tools.
- Write a label policy that says when a visible disclosure is required.
- Ask each AI vendor how it marks outputs and whether the marks survive compression and editing.
- Keep a human editor in the review loop and record who approved each asset.
- Set up a response plan for assets that are challenged or mislabeled.
Illustrative label policy (example only)
- Fully AI-generated image or video: visible label plus preserved metadata.
- AI-assisted draft with human rewrite and editor sign-off: internal record, public label per channel rules.
- Synthetic voice or likeness of a real person: written consent and a visible label.
Laws and guidance in this area are changing quickly, so confirm current status before acting.
Frequently asked questions
What is AI disclosure?
It is the practice of marking content as AI-generated or AI-assisted so people, platforms and regulators can identify it, using both visible labels and embedded machine-readable marks.
Is AI disclosure legally required?
In some cases, yes. The EU AI Act's Article 50 and California's AI Transparency Act both apply from August 2, 2026, but each covers specific actors and content types, so coverage depends on the facts.
Does a label on the content satisfy the law?
Not by itself. California's law requires a machine-readable latent disclosure for covered providers, and Article 50 requires machine-readable marking of generated content, so a caption alone does not cover the provider duty.
Who is responsible, the brand or the AI vendor?
Often both. Providers owe the machine-readable marking of their outputs, while deployers, including brands, may owe labels for content such as deepfakes. A vendor contract should state who handles each duty.
Does text need an AI label?
California's law covers image, video and audio. The EU rules reach AI-generated text only in specific cases, such as text published to inform the public on matters of public interest, and an editorial-control exception may apply.
By the 5W Editorial Team. Reviewed by a human editor before publication. Updated October 28, 2026.