# California SB 1050: AI Ad Disclosure Rules for Advertisers

> California's SB 1050 requires AI ad disclosure for synthetic performers from Jan 1, 2027. What it covers, how it differs from New York, and how to comply.

[← Back to Blog](/blog/)

Strategy September 21, 2026 8 min read

# California SB 1050: AI Ad Disclosure Rules for Advertisers

California just became the second state to require a disclosure when an ad prominently features an AI-generated performer. Adding the label is trivial. Knowing which of your ads need it — across hundreds of AI-generated variants — is the actual compliance project, and it starts before January 1, 2027.

On September 16, Governor Newsom signed **SB 1050**, California's new **AI ad disclosure law**. It requires a clear and conspicuous disclosure on any video or audio advertisement that prominently features a "synthetic performer" — an AI-generated human who is not a real, identifiable person. The bill was sponsored by SAG-AFTRA, and the Governor signed it at the union's headquarters. It takes effect **January 1, 2027**.

Most coverage has framed this as a performers'-rights story, which it is. For anyone who buys media, it is something more mundane and more demanding: a record-keeping requirement disguised as a labeling requirement. This post covers what the law actually says, how it differs from New York's version that took effect in June, and where it will bite teams that produce AI creative at volume.

## What SB 1050 Actually Requires

The operative rule, per the [Davis+Gilbert summary of the enacted text](https://www.dglaw.com/california-becomes-second-state-to-require-disclosure-of-synthetic-performers-in-advertising/): it is unlawful to create and cause to be published an advertisement that _prominently_ includes a synthetic performer without a clear and conspicuous disclosure.

### Who counts as a synthetic performer

A digital figure, voice, or representation created in whole or in part with generative AI that creates the realistic impression of a human performer who is _not_ an identifiable natural person. Two things follow from that definition. Voices count, not just faces. And AI replicas of real, identifiable people are outside this law — California already regulates digital replicas of real performers through [separate 2024 legislation](https://www.gov.ca.gov/2024/09/17/governor-newsom-signs-bills-to-protect-digital-likeness-of-performers/).

### When a synthetic performer is "prominent"

The disclosure is triggered only when the synthetic performer:

-   appears in the foreground demonstrating or illustrating the product or service;
-   provides or voices the on- or off-camera narration or commercial message; or
-   illustrates or reacts to that narration or message.

### What the disclosure must say

Wording "substantially similar" to **"This performance features a synthetic performer"** or **"No human performer is depicted."** It must be "difficult to miss, easily understandable, and presented in a manner that a reasonable consumer would notice, read, and comprehend."

### What is exempt

Ads for expressive works — films, TV, streaming content, video games — where the synthetic performer is used consistently with the underlying work. And generative AI used solely for language translation, so dubbing a real actor into Spanish does not trigger it.

### Enforcement

A violation is treated as a violation of California's false advertising statute (Business & Professions Code § 17500) and is enforceable through the Unfair Competition Law. Both [Kelley Drye](https://www.kelleydrye.com/viewpoints/blogs/ad-law-access/california-enacts-a-synthetic-performer-law) and Davis+Gilbert flag that this route potentially opens the door to private lawsuits — a meaningful change from New York. Separately, once a court finds an ad non-compliant, the platform or publisher running it must pull it as soon as commercially reasonable.

## California vs. New York: Build to the Union of Both

New York's synthetic performer law [took effect June 9, 2026](https://www.kelleydrye.com/viewpoints/blogs/ad-law-access/ny-synthetic-performer-law-goes-into-effect). The two laws overlap but are not interchangeable:

| Question | New York | California (SB 1050) |
| --- | --- | --- |
| Which uses trigger disclosure? | Arguably any synthetic performer, including extras | Only "prominent" ones |
| Audio-only ads? | Exempt | Covered |
| Required wording? | None specified | Substantially similar to prescribed phrases |
| Knowledge requirement? | "Actual knowledge" | No such qualifier |
| Enforcement | Civil penalty $1,000 first / $5,000 subsequent; no private right of action | False advertising + UCL; potential private suits |
| Effective | June 9, 2026 | January 1, 2027 |

A national campaign cannot practically serve one creative to New York and another to California, so the working rule is to build to the stricter side of each row. That means New York's broad trigger (disclose background and partial synthetic humans too, as Kelley Drye notes many advertisers already do conservatively), California's prescribed wording (which is also a reasonable answer to New York's unspecified wording), and California's coverage of audio.

## The Case That This Is a Non-Event

Before planning a compliance program, it is worth taking the dismissive read seriously, because most of it holds.

The disclosure costs almost nothing. It is one line of on-screen text or a few seconds of spoken audio. Most brand-safe advertisers were not running photorealistic fake humans as spokespeople to begin with. The prominence test in California narrows the scope further than New York's. Nobody will be sued over a blurred AI-generated crowd behind a real actor. And the effective date is more than three months away — ample time to add a line to an end card.

There is also a reasonable argument that the label changes little for consumers. Whether audiences penalize disclosed AI creative is an open empirical question, and the answer probably varies by category. If disclosure does not hurt performance, the law is a formatting rule.

All of that is correct about the _label_. It is wrong about what it takes to know where the label goes.

## Where It Actually Bites: The Inventory Problem

The disclosure is a one-line fix. The obligation it creates is to know, for every live ad, whether a generative model produced a realistic human in it — and California's version contains no "actual knowledge" escape hatch. "Our tool generated that and nobody noticed" is not a defense the statute provides.

That matters because of how performance creative is now made. Teams running [AI creative testing at scale](/blog/ai-creative-testing-at-scale.md) produce dozens or hundreds of variants per concept. Some of those variants come from avatar tools, some from image-to-video features, some from platform-side generation. Several of these generate or alter humans without anyone on the account explicitly choosing to.

### AI voiceover is the sleeper exposure

The prominence test includes whoever "provides or voices the on- or off-camera narration." A realistic text-to-speech voice reading your script is plausibly a synthetic performer delivering the commercial message — and in California, audio-only ads are covered. AI voiceover is far more common in performance video than AI avatars, and it rarely gets flagged as "AI creative" internally because nobody sees it. If your team uses TTS for UGC-style ads, product demos, or streaming audio spots, that is the first inventory to take.

### Platform-generated variants

Platforms increasingly generate or transform creative on your behalf — the [Performance Max video reframing opt-out](/blog/performance-max-ai-video-reframing-2026.md) we covered last month is one example, and Meta's generative features inside [Advantage+](/blog/meta-advantage-plus-2026-ai.md) are another. Most of these adjust formats, backgrounds, or motion rather than invent people, and the law applies to whoever "creates and causes" the ad to be published. But the safe default is to know which auto-generation features are switched on for each account, rather than to assume none of them produce a human.

### Platform AI labels are not a substitute

Several platforms apply their own AI-content labels. Those are platform policies, not the statutory disclosure, and they do not use the prescribed wording. Until counsel tells you otherwise, do not assume a platform badge satisfies SB 1050.

## A Practical Compliance Checklist Before January 1, 2027

### 1\. Inventory every live ad by performer type

For each ad currently running or scheduled past January 1: real human, AI-generated human (visual), AI-generated voice, or none. Include evergreen assets that have been running since last year — they are covered the day the law takes effect.

### 2\. Add a provenance field at the point of creation

The cheap time to record "contains synthetic performer: yes/no" is when the asset is generated, not six months later when someone asks. Put it in the asset naming convention, the DAM metadata, or the ad name itself. Kelley Drye's advice is the same in plainer terms: start cataloging which ads include synthetic performers, and make your agencies do it too.

### 3\. Standardize one disclosure and reuse it everywhere

Pick the prescribed phrase, define minimum on-screen duration, size, and contrast for video, and a spoken version for audio. One template applied consistently is easier to defend than per-campaign improvisation.

### 4\. Put it in agency and creator contracts

The law reaches whoever creates and causes the ad to be published. If an agency or UGC creator delivers an asset, require them to declare synthetic performers at delivery.

### 5\. Treat automation as a creator, not a bystander

If an AI agent or script generates and uploads ads, it should write the provenance flag and attach the disclosure as part of the same step. Anything that can launch creative across accounts — the kind of [multi-platform orchestration](/blog/multi-platform-ai-ad-orchestration.md) agencies are building now — is also the right place to enforce the rule automatically.

### 6\. Get legal review of the edge cases

This post is not legal advice. Partial performers, stylized-but-realistic characters, and AI-enhanced footage of real actors all sit in gray areas neither state has clarified. Your counsel should decide where your line is before a plaintiff does.

## The Takeaway

SB 1050 will not change what a well-run ad looks like; one line of text is not a creative constraint. What it changes is the bookkeeping. Two large states now require you to know which of your ads contain AI-generated people, one of them without regard to whether you knew — and more are likely to follow.

Teams that treat this as a labeling task will scramble in December, auditing hundreds of variants by eye. Teams that treat it as a data problem — record provenance when the asset is made, carry it through to the ad, attach the disclosure automatically — will barely notice the law took effect.

_Sources: [Office of the Governor — Governor Newsom signs new law to protect workers, require disclosures on AI-generated advertising (September 16, 2026)](https://www.gov.ca.gov/2026/09/16/governor-newsom-signs-new-law-to-protect-workers-require-disclosures-on-ai-generated-advertising/), [Kelley Drye Ad Law Access — California Enacts a "Synthetic Performer" Law](https://www.kelleydrye.com/viewpoints/blogs/ad-law-access/california-enacts-a-synthetic-performer-law), [Davis+Gilbert — California Becomes Second State to Require Disclosure of Synthetic Performers in Advertising](https://www.dglaw.com/california-becomes-second-state-to-require-disclosure-of-synthetic-performers-in-advertising/), [Kelley Drye — NY "Synthetic Performer" Law Goes into Effect](https://www.kelleydrye.com/viewpoints/blogs/ad-law-access/ny-synthetic-performer-law-goes-into-effect)_

### Make provenance part of the launch step

Ads Agents connects your AI agents to Google, Meta, and TikTok through MCP — so the same workflow that generates and launches creative can tag it, audit it, and pause what needs fixing across every account.

[Get Started Free →](https://app.ads-agents.com/registration)