Do You Have to Label AI-Generated Ads? The Q4 2026 Rules Map

You have to label AI in your own ads only in a few places, and each law means something different by "AI". New York has required a conspicuous label on AI-generated people in ads since 9 June 2026. California signed a version on 16 September 2026 that also covers AI voices. The EU has required a visible label on deepfakes since 2 August 2026. Origin Pi found no general labelling rule in the UK, Singapore, Australia or US federal law, but misleading-ad law applies everywhere and platforms set their own label rules. It is for marketing and compliance leads running one campaign in several markets.

Update, 28 October 2026: We re-checked the TikTok, Google and Meta ad policies, the EU Code of Practice page and New York's GBL 396-b text. None has changed, and California's SB 1050 start date is still our reading rather than a confirmed date; the rest of the map stands as of 17 October.

Rules as of 17 October 2026. New: on 16 September 2026 the governor signed SB 1050, so two large US states now each have a synthetic-performer law.

How we checked: Origin Pi read official texts or regulator pages for the 10 matrix markets in October 2026. Out of scope: other US states, political-ad and sector rules. India is covered by one note below.

Which ads need an AI label, market by market?

Three AI-label duties bind advertisers on 17 October 2026: AI-made human performers in New York ads, realistic deepfakes reaching the EU, and AI content posted in mainland China. California signed a fourth on 16 September 2026. Elsewhere, the test is whether leaving the label off would mislead.

This is general information, not legal advice. Ask counsel where a call is close.

Three rows need a word. China's labelling measures (our translation), in force since 1 September 2025, tell users who publish AI content to declare it and use the platform's label tool (Article 10); the regulator acted against two apps and a website on 28 April 2026. South Korea's AI Basic Act, in force since 22 January 2026, puts the label on the AI business operator (Article 31), with no fines for at least a year per the science ministry. Malaysia's online safety code, enforced since 1 June 2026, makes licensed platforms label realistic AI media. India sits outside the matrix. The Ministry of Electronics and IT says in a PIB release of 6 August 2026 that on 10 February 2026 the government amended the IT Rules 2021 so that intermediaries must label AI-generated content and attach traceable metadata. Check MeitY's notification for what applies to your ads.

US scope: Origin Pi found no federal statute requiring AI labels on ads as of 17 October 2026 (pending bills are not law). Federally, section 5 of the FTC Act bars unfair or deceptive practices, enforced through FTC orders (see the Active Listening case below). New York 396-b and California SB 1050 are state law. Political-ad and sector rules are out of scope.

What do New York and California require for AI people in ads?

New York and California both require a clear disclosure when an ad features an AI-generated human performer who is not a recognisable real person. New York's General Business Law 396-b has applied since 9 June 2026. California's SB 1050, signed on 16 September 2026, adds AI voices: the governor's release says it requires explicit disclosure on any video or audio advertisement that uses AI-generated performers. Our reading: California statutes usually take effect on 1 January after enactment, so we expect it to apply from 1 January 2027; check the chaptered bill.

New York's law covers AI-made human performers who are not recognisable real people, and the advertiser must disclose where it has "actual knowledge". The enacted bill text sets a civil penalty for each violation, higher after the first.

"shall conspicuously disclose in such advertisement that a synthetic performer is in such advertisement"

New York General Business Law 396-b(3), nysenate.gov, law signed 11 December 2025, applying from 9 June 2026.

California is narrower in one way and wider in another. It bites only when the synthetic performer appears "prominently". Per the Assembly committee analysis (June 2026, pre-final), its definition covers a human-like digital figure, voice or representation, so an AI voice counts. Suggested wording: "this performance features a synthetic digital performer".

The analysis says a breach is a violation of California's false-advertising law, enforceable under the unfair competition law, and that media must pull an ad a court has found in breach. The analysis states no criminal or per-violation civil penalty for SB 1050, so this post states none.

A separate California law sits on the other side of the ad. The AI Transparency Act became operative on 2 August 2026 ("Beginning August 2, 2026", per the Assembly committee analysis) and makes generative AI providers embed hidden provenance data in images, video and audio: a tool-maker duty. The Attorney General, a city attorney or a county counsel can seek a civil penalty per violation, and each day in breach counts as a separate violation.

Illustrative example: a skincare brand runs one video in New York, California and Texas, with an AI presenter who resembles no real person and an AI voice. New York needs a conspicuous disclosure today. California will too from 1 January 2027, when SB 1050 starts, including on a radio cut, which New York exempts. Origin Pi found no Texas statute, so platform rules decide there.

How does the EU rule differ, and who carries it?

The EU splits the job between two parties. Article 50(2) of the AI Act makes the tool provider mark its output "in a machine-readable format". Article 50(4) makes the deployer (the publisher) disclose deepfakes, and has applied since 2 August 2026. For systems placed on the market before then, the Digital Omnibus (Regulation (EU) 2026/1744) gives providers until 2 December 2026 for Article 50(2); new systems have no grace.

Breaches can be fined up to 3% of worldwide turnover or a fixed cap, whichever is higher, and whichever is lower for SMEs.

A deepfake under Article 3(60) is content that "resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful", so a misleading AI product shot counts. The Commission's Article 50 guidelines of July 2026 (paragraph 116) treat aesthetic background replacement as likely minor, but an image that makes the product look better than it is as a deepfake, case by case. See our guide to the 2 December deadline.

The Commission's Code of Practice on marking and labelling (10 June 2026) is voluntary: signatory tool makers commit to signed metadata plus an imperceptible watermark, and signatory deployers to an "AI" icon or equivalent label.

What do Meta, Google, TikTok and YouTube require?

TikTok has the strictest platform rule. Its ads policy, last updated April 2026, says "If we identify AI-generated content that has not been disclosed, your ad will be rejected or restricted." These are policies, not law.

What each platform asks of an advertiser running AI creative:

Google's July 2026 change is not a new global mandate. Its policy change-log of 9 July 2026 says Google now "permits" AI labels inside ad creatives and that the setting "doesn't guarantee compliance". Meta's 1 June 2026 update says it will "begin automatically detecting ads created or edited using third-party AI tools", so an unlabelled AI ad may get a label anyway.

Is a label enough? What the FTC's Active Listening orders show

A label is unlikely to rescue an ad that misleads. The US Federal Trade Commission's August 2026 orders concerned a false claim about AI, not a missing label, and the UK and Singapore watchdogs agree.

The FTC announced final orders on 27 August 2026 against Cox Media Group, MindSift and 1010 Digital Works. They sold an "Active Listening" service said to use "a special algorithm to listen in on and detect pertinent conversations from smart devices". The FTC says it was not based on voice data. This is AI-washing: overstating what AI does.

The FTC acted under the FTC Act, which bars deceptive practices. The lesson holds for any AI claim: say only what the product does.

The UK's advertising regulator has no AI disclosure rule and sees no cure in one: its May 2025 guidance says "There is no blanket legal requirement in the UK to disclose the use of AI in ads". Singapore's ASAS, in its 23 March 2026 release, called "indiscriminate and vague disclosures of AI usage" unhelpful, and the ministry's 5 August 2026 written answer said it will "assess if they should be mandated". See also who is liable for AI ad claims.

"disclosure alone is very unlikely to mitigate the harm caused by a fundamentally misleading message"

UK ASA and CAP, guidance on disclosing AI in advertising, 29 May 2025.

How do you check a campaign before it ships?

A campaign team can clear AI labelling with three checks per asset: what the AI made, where the ad runs, and what the ad claims. Run the decision tool once per market.

Test yourself on three cases campaign teams meet.

Our free AI compliance starter pack has the asset log and approval templates.

Where Origin Pi stands

We think most teams over-label the wrong assets and under-check the right ones. The bigger exposure is a realistic AI image or claim that misleads, which a label is unlikely to cure.

We design AI compliance workflows that record which tool made each asset, whether it shows a person or a voice, and which markets it targets, and hold risky assets until a named person approves. For marketing teams shipping AI creative at volume, that is cheaper than relabelling after launch.

Your first step this week: list every AI-made person and voice in the creative running in New York now, and check each against the decision tool.

Take the data with you

Every figure here is in one table with sources; the card is shareable. Please cite this post if you reuse either file.

• Data table (CSV, 4 KB)
• At-a-glance card (PNG, 84 KB)