gotalt?

AI disclosure laws: which rules ask a business to label AI content

There is no single AI disclosure law. Article 50 of the EU AI Act has applied since August 2, 2026 and covers chatbots, AI-generated media and some AI-written text. In the United States no federal statute requires labeling, and the state rules are narrow: AI-generated people in ads, altered real estate photos and chatbots. Platforms add rules of their own. Each rule below shows who it binds, its date and the primary source we read.

Published . Last reviewed .

This page maps the AI disclosure rules we have read in their primary texts and links each one to a page that quotes it. It is general information, not legal advice: whether a rule applies depends on facts about you, such as where your customers are, who built your chatbot and what your images show. To get a first list for your own situation, answer a few questions in which AI disclosure rules apply to me. For accessibility law, see web accessibility laws by country.

The main rules at a glance

Dates are written as they appear in the source. "Last verified" is the date we last checked the row against its primary source. A platform rule is a contract term, not a law, and the table says so.

AI disclosure rules by jurisdiction, with who they bind, key dates and where to read more
Where Rule Who it binds What it asks Status and dates Last verified Read more
European Union AI Act Article 50(1): chatbots and other AI that talks with people Providers of AI systems that interact directly with people. A company that supplies a chatbot under its own name or trademark is the provider. Design the system so people are told they are interacting with an AI system, unless that is obvious. Applies since August 2, 2026, with no transition period. EU: chatbots
European Union AI Act Article 50(2): marking AI output Providers of AI systems that generate synthetic audio, image, video or text Mark output in a machine-readable format so it is detectable as artificially generated or manipulated. Applies from August 2, 2026. Providers of generative systems placed on the market before that date have until December 2, 2026. EU: marking
European Union AI Act Article 50(4): deep fakes and AI text on matters of public interest Deployers: businesses and professionals that use AI to generate or manipulate content Disclose that a deep fake is artificially generated or manipulated. Label AI text published to inform the public on matters of public interest, unless it was human-reviewed and someone holds editorial responsibility. Applies since August 2, 2026. Content generated before that date does not need to be labeled retroactively. EU: deep fakes
European Union AI Act Article 50(5): how the information is given Everyone bound by Article 50(1) to (4) Provide the information clearly and distinguishably by the first interaction or exposure, in a way that conforms to the applicable accessibility requirements. Applies since August 2, 2026. EU: accessibility
European Union Code of Practice on Transparency of AI-generated Content Signatories only Voluntary. Section 2 covers deployers: the EU icon or an equivalent label, its placement and accessibility measures. Final text June 10, 2026. The Commission found it adequate on July 8, 2026. EU: Code and icons
New York General Business Law 396-b: synthetic performers in advertising A person in the business of dealing in property or services who produces or creates an advertisement, where that person has actual knowledge Conspicuously disclose in the advertisement that a synthetic performer is in it. In force since June 9, 2026. $1,000 for a first violation and $5,000 for each later one. US: New York
California SB 1050 (Business and Professions Code 17610): synthetic performers in ads Any person who creates and causes publication of an ad that prominently includes a synthetic performer Disclose it, in wording substantially similar to "this performance features a synthetic performer" or "no human performer is depicted". Signed September 16, 2026. Effective January 1, 2027. US: SB 1050
California AB 723 (Business and Professions Code 10140.8): altered real estate images Real estate brokers, salespeople and persons acting on their behalf State that the image has been altered, with a link or QR code to the original, on or next to the image. In force since January 1, 2026. A willful violation is a misdemeanor (up to $10,000 or six months). US: AB 723
California, Maine, Utah Chatbot disclosure: California SB 1001 and AB 1609, Maine 10 MRSA 1500-Y, Utah Code 13-75-103 and 13-75-104 The business that uses the chatbot. California AB 1609 binds only businesses with more than $500,000,000 in annual revenue. Notify the consumer that it is not a human where it could mislead. Utah offers a safe harbor for disclosing at the outset and throughout. California SB 1001 since July 1, 2019. Utah since May 7, 2025. Maine's act was approved June 12, 2025. California AB 1609 effective January 1, 2027. US: chatbots
California AI Transparency Act (Business and Professions Code 22757.1 and following) Providers of generative AI systems publicly accessible in California and large online platforms. Not ordinary sellers. Embed latent disclosures and offer a free verification tool (providers); surface provenance data and not knowingly strip it (large platforms). Operative August 2, 2026. Rewritten by SB 1000 on September 30, 2026. US: provenance laws
Illinois, Connecticut, Colorado Notice duties for employers, AI subscriptions and consequential decisions Employers using AI in employment decisions (Illinois); sellers of AI subscriptions to Connecticut consumers; deployers of automated decision technology in consequential decisions (Colorado) Give the notice each law describes. Illinois since January 1, 2026. Connecticut from October 1, 2026. Colorado from January 1, 2027. US: notice duties
Texas Responsible Artificial Intelligence Governance Act (HB 149) Disclosure duties bind government agencies and health care providers, not private businesses in general No consumer AI disclosure duty on private businesses. In force since January 1, 2026. US: Texas
United States (federal) No AI labeling statute or FTC rule. Deception law applies. Any business, under FTC Act section 5 and the Consumer Reviews and Testimonials Rule No AI label. Claims about AI products need evidence, and fake reviews are banned however they are made. No AI labeling rule in force. US: FTC
China Measures for Labeling AI-Generated Synthetic Content Service providers, content distribution platforms and users who publish AI-generated content on them Explicit labels people can perceive, implicit labels in file metadata and a user declaration on platforms. In force since September 1, 2025. Enforcement action announced April 28, 2026. China
South Korea AI Basic Act, Article 31 AI business operators that provide generative AI or products or services using it Notify users in advance, indicate that output was generated by AI and mark hard-to-distinguish synthetic output. In force since January 22, 2026. Secondary sources report fines are suspended for at least a year. South Korea
Platforms Google Merchant Center, Google Ads, YouTube, TikTok, Meta, Amazon, Etsy and others Sellers, advertisers and creators who use the platform Disclosures at upload, tags in image files, or labels in listings. Each platform differs. Platform rules, not laws. Dates are on the platforms page. Platform rules

Who the duty sits on

The same image can raise different duties for different people, so the first question is the role, not the technology.

  • Provider: a person or body that develops an AI system, or has one developed, and puts it on the market or into service under its own name or trademark. The EU's chatbot duty (Article 50(1)) and marking duty (Article 50(2)) sit here.
  • Deployer: a person or body that uses an AI system under its own authority, other than in a purely personal, non-professional activity. The EU's deep fake and AI text duties (Article 50(4)) sit here.
  • Advertiser: the business that creates an ad. New York's and California's synthetic performer laws sit here.
  • Business using a chatbot: California SB 1001, Maine and Utah put the notice duty on the business that uses the bot, not on the vendor that built it.
  • Real estate licensee: California AB 723 binds licensees and people acting on their behalf, which may include a virtual staging vendor or photographer working for a licensee.
  • Seller, creator or advertiser on a platform: platform rules bind whoever uploads. They are terms of service, not laws.

The chatbot duty is the clearest case of why this matters. In the EU, the Commission's Guidelines say a company that supplies a chatbot under its own name or trademark is the provider, including when customers use it without modification, so a store that embeds an unmodified vendor widget is most likely a deployer. In Maine, Utah and California the duty sits on the business that deploys the bot. Any checklist that treats the two as the same is wrong for one of them.

Common situations and the rules that may apply

These are pointers to the rules that may apply, not conclusions. Each links to the page that quotes the text.

Common situations, the AI disclosure rules that may apply and where to read them
Your situation Rules that may apply Read
You use AI to make product images for your own store EU Article 50(4) if an image is a deep fake. The Guidelines list an AI-generated product image that can mislead about the real product as a deep fake, and a real product shown against an AI-generated background as not one, as long as the ad is not likely to mislead. Google Merchant Center asks for the IPTC DigitalSourceType value trainedAlgorithmicMedia on AI-generated product images. EU: deep fakes, Google Merchant Center, AI image metadata
Your ads show AI-generated people or voices New York 396-b (visual or audiovisual ads, actual knowledge), California SB 1050 from January 1, 2027 (includes audio), the EU deep fake rules, and Amazon's reported keyword for photorealistic AI-generated people. US: ads, Synthetic performer ads
You publish real estate photos that software or AI has altered California AB 723 for licensees and people acting for them. The Commission's icons page lists an authentic photograph of an empty apartment furnished using AI among the uses of its "Partially AI-Modified" icon. US: AB 723, Virtual staging disclosure
You run a chatbot or AI agent EU Article 50(1), where the provider carries the duty; California SB 1001, Maine and Utah, where the business that uses the bot does; California AB 1609 for businesses over $500,000,000 in revenue. EU: chatbots, US: chatbots, AI chatbot disclosure
You publish AI-written or AI-edited text EU Article 50(4) applies only to text published to inform the public on matters of public interest. Product descriptions and ad copy are outside it unless they carry claims about health, consumer safety or sustainability. No US state rule for publishers is in force; a New York bill has passed both houses but has not been delivered to the Governor. EU: AI text, US: pending bills
You want your AI label to work with a screen reader EU Article 50(5) points to the accessibility rules that already apply. The Code of Practice lists accessibility measures for signatories. A label drawn only into the pixels of an image is invisible to a screen reader. EU: accessibility, Accessible AI labels, Label generator
You sell or build a generative AI tool for others EU Article 50(1) and (2) as provider, the California AI Transparency Act and similar provenance laws in Connecticut, Utah and Washington, with user thresholds that differ by state. EU: marking, US: provenance laws

Claims we do not repeat

Several statements about these rules circulate that the primary texts do not support. We leave them out, and say why.

  • "The Digital Omnibus delayed Article 50." It did not, except for one duty. Regulation (EU) 2026/1744 gives providers of generative AI systems placed on the market before August 2, 2026 until December 2, 2026 to meet the marking duty in Article 50(2). Chatbot disclosure and deep fake labeling apply from August 2, 2026 (Guidelines, paragraph 153).
  • "A metadata tag is the EU label for a deep fake." The Commission says deployers cannot rely on the machine-readable marking the provider embeds, because people cannot perceive it (Guidelines, paragraph 117).
  • "Using the EU icon shows you meet the Act." The Commission's icons page says use of the icons "does not establish legal compliance by itself", and that a non-signatory's use should not be read as adherence to the Code.
  • "New York's 396-b is a general deepfake label law." The enacted text is narrow: advertisements, synthetic performers who are not recognizable as an identifiable person, and the business that creates the ad, with actual knowledge. A sponsor memo describes something broader; the statute does not.
  • "AI-generated alt text must be marked." The Guidelines list alt text, image captions and UI labels among short outputs outside the marking duty in Article 50(2) (paragraph 68).
  • "Retroactive tagging of old Amazon images is required." We could not read Amazon's own help page, which is behind a seller login, and the reports we found disagree on older images. We do not say it is required.

Enforcement so far

As of October 21, 2026 the only enforcement of an AI labeling rule that we could confirm from a primary source is China's. On April 28, 2026 the Cyberspace Administration of China named the CapCut (Jianying) app, the Maoxiang app and the Jimeng AI website for failing to label AI-generated content, and ordered interviews, corrections and warnings. Its notice states no fine (details).

We found no enforcement action under EU AI Act Article 50, New York General Business Law 396-b, California AB 723 or the California AI Transparency Act, or under Korea's AI Basic Act. The FTC has acted on claims made about AI products: the final order against accessiBe (April 22, 2025, $1,000,000) bars claims, without evidence, that an automated product makes a website meet WCAG, and the final order against Workado (August 28, 2025) requires competent and reliable evidence for accuracy claims about an AI content detector. Neither is a labeling case, because the FTC has no AI labeling rule to enforce.

Our research had limits. There is no central register of EU enforcement, and several official sites blocked automated reads. Not finding an action is not proof that none exists.

Sources: Cyberspace Administration of China, April 28, 2026 (Chinese text); FTC, accessiBe final order; FTC, Workado final order. Verified .

Free tools and guides

Each tool runs in your browser and works from what a file or page declares. None of them decides whether an image is AI-generated, and a result of "nothing found" proves nothing about how an image was made.

Guides: accessible AI labels, alt text for AI-generated images, virtual staging disclosure, AI chatbot disclosure, synthetic performer ads and AI image metadata.

What we don't claim

This page is general information, not legal advice. No tool, ours included, can confirm that a site, an image or a chatbot meets any of these rules, and GotAlt does not sell a badge or seal that says so. We do not detect whether an image is AI-generated: a missing marker proves nothing, and a present one is a declaration, not proof. The Commission's Guidelines are non-binding, and several of these laws changed in the weeks before this review, so check the date on each row. We have not covered Mexico, most of Latin America, the UK's fake-review rules, EU political advertising rules or which marketplaces the Digital Services Act designates. Our editorial policy explains how we source and date law pages.

AI disclosure laws: common questions

Is there one law that requires businesses to label AI content?

No. The closest is Article 50 of the EU AI Act, which has applied since August 2, 2026 and asks different things of providers and deployers. In the United States we found no federal statute or FTC rule that requires AI labels, and the state rules are narrow, each with its own trigger. Platform rules add duties that are contract terms, not laws.

Does the EU AI Act reach a business outside the EU?

It can. The Commission's Guidelines say a deployer outside the EU is bound where it foresees dissemination and use of the AI output in the Union, including by posting deep fakes on the globally accessible internet, and is not bound where content reaches EU audiences through channels that are unforeseeable and outside its control (paragraph 13). The Guidelines are non-binding, and only the Court of Justice interprets the Act.

Do I have to label every AI-edited photo?

Not under Article 50(4). That duty attaches to a deep fake: content that resembles existing persons, objects, places or events and would falsely appear authentic. For product advertising, the Guidelines expect AI color correction, background extensions, background replacement for clearly aesthetic purposes, arrangement of existing products and rescaling to have only a minor impact (paragraph 116). State and platform rules use different tests.

Does an AI image need a metadata tag, a visible label or both?

They do different jobs. For an EU deep fake the label must be perceivable by people, and deployers cannot rely on the machine-readable marking a provider embeds (paragraph 117). Platforms such as Google Merchant Center read a tag in the image file. A visible label does not replace a tag a platform asks for, and a tag does not replace a visible label.

Has anyone been fined for missing an AI label?

We found no fine under EU Article 50 or under a US state disclosure law. China's regulator took action against three named services on April 28, 2026, ordering interviews, corrections and warnings, and its notice states no fine. Our research cannot rule out actions we did not find.

Can a badge or icon show that my site meets these rules?

No. The Commission says use of its icons does not establish legal compliance by itself, and Google says its AI label setting does not guarantee compliance. GotAlt does not sell a badge or certificate.

See what your pages declare and show

Check up to 50 pages once, $49, no subscription. The Site Image Check includes an AI disclosure section: for each image, the AI markers the file declares and whether the page shows AI-label text with it; for each page, known chat widgets and whether AI-disclosure wording appears in the page HTML. It never decides whether an image is AI-generated. To test one page first, use the free AI label checker.

Check my site once from $49