Correction, 2026-10-02: we re-checked every source on this page. The IAB figures stand. Several legal and platform points were looser than the sources. The FTC's 2023 update changed the definition of endorsements "to clarify the extent to which it includes" virtual influencers, which is narrower than our "covers virtual influencers". Google's rule is a checkbox disclosure for election ads with synthetic or digitally altered content, plus a prominent in-ad disclosure on some formats. Our FAQ answer on labels left out TikTok's ad policy (a label for completely AI-generated content), New York's synthetic performer disclosure law (in effect since June 9, 2026), California's (from January 1, 2027) and the EU AI Act's deep fake rule. We also said a synthetic person shown using and praising a product was fine if the experience matched reality; the FTC's reviews and testimonials rule targets testimonials by someone who does not exist or who had no actual experience with the product, so an AI person should present, not testify. We removed unsourced statements about what "most teams" and "the teams handling this well" do, replaced a "locked source" that could read as a lock feature 8frame does not have, added the plans needed to invite reviewers, and added a note that this page is general information, not legal advice.
Correction, 2026-10-01: the first version of this page said 40% of Gen Z and Millennial consumers feel negative about AI ads. IAB's study gives no overall negative share; it reports that the share feeling somewhat or very negative is 12 points higher than in 2024. It also claimed, from "the 8frame team's read across client accounts", that a "Made with AI" label does not measurably reduce click-through; we have no such data, so that is removed and replaced with IAB's survey finding. The platform disclosure paragraph overstated the rules: Meta and Google require disclosure of realistic synthetic content in political, social-issue and election ads, not in every ad. Sources are now listed.
AI brand safety in 2026 is two problems wearing one name. The first is the safety of the creative itself: a generated ad can invent a product claim, drift a logo, put words in a synthetic spokesperson's mouth, or render a competitor's trademark by accident. The second is the safety of the placement: an ad you know is AI-made may need an AI label, and it runs in front of consumers whose feelings about AI ads are measurably mixed. The placement half includes the familiar adjacency problem that brand-safety controls were built for. The creative half, the one that can put a false claim in a paid unit at scale, is the one that needs a new QA gate. This piece covers both, with a review workflow you can run on every AI-generated asset.
This is general information, not legal advice. Laws and platform policies change, and how they apply depends on your ad and where it runs. Every rule below is linked in Sources (checked 2026-10-02); ask a lawyer about your specific campaign.
Why this got harder in 2026
The volume changed. When a brand shipped two or three hero creatives a quarter, a single human review pass could catch most problems. When the same brand ships 40 variants a week, the review surface multiplies and so do the failure modes. In IAB and Sonata Insights' "The AI Ad Gap Widens" study (January 15, 2026), 83% of US ad executives said their company has deployed AI in the creative process, up from 60% in the 2024 study. That is a lot of new output moving through pipelines that were designed for hand-made assets.
The exposure also changed. The same IAB study found that 71% of Gen Z and Millennial consumers believe they have seen an ad created using AI, up from 54% in 2024, and that the share who feel somewhat or very negative about the use of generative AI to create ads is 12 points greater than it was in 2024. Consumers are watching for AI, so a slip can read as a brand that let a machine run unsupervised rather than as an ordinary production error. Brand safety is no longer only about where the ad appears. It is also about whether the ad can defend itself.
Half one: safety OF the creative
This is the newer half. Four failure modes cover most of what a review needs to catch.
Hallucinated claims. A prompt like "energetic ad for our protein bar, highlight the benefits" invites the model to invent benefits. Generated copy and voiceover can assert "clinically proven," "#1 dermatologist recommended," or a specific gram count the product does not have. Every factual claim in an AI-assisted asset needs to trace back to something the brand can substantiate. Treat generated copy as a first draft from an enthusiastic intern who has never read the product spec sheet.
Likeness and identity drift. If your creative features a synthetic presenter or a licensed avatar, the risk is that the face shifts across variants or ends up resembling a real person you have no rights to. For the consent and licensing rules on real people's likeness, see synthetic talent and celebrity likeness in ads.
Endorsement rules apply here too. The FTC's 2023 update to its Endorsement Guides changed the definition of endorsements "to clarify the extent to which it includes fake reviews, virtual influencers, and tags in social media," and the FTC's guidance says an endorsement "must reflect the honest opinion of the endorser and can't be used to make a claim the marketer of the product couldn't legally make." Separately, the FTC's rule on consumer reviews and testimonials, effective October 21, 2024, "addresses reviews and testimonials that misrepresent that they are by someone who does not exist, such as AI-generated fake reviews, or who did not have actual experience with the business or its products or services." So a synthetic person can present your product and deliver claims that are true and substantiated, but it should not pose as a customer describing an experience it never had. Where that line falls for a given script is covered in do you have to disclose AI-generated ads; ask counsel about anything in between.
Trademark and logo drift. Image and video models redraw a logo from a reference; they do not paste the exact file. Across a 30-variant batch, a wordmark can lose a letter, shift color, or pick up an artifact that is invisible at thumbnail size and obvious on a CTV screen. The same drift can pull a competitor's mark or a recognizable third-party brand into the background of a scene. Add brand marks as overlays in your editor rather than trusting the model to render them, and scan backgrounds for unintended trademarks.
Off-brand tone and unsafe context. Loose direction can produce generic results: a stock-looking family, a cliche office, a sunset that belongs to no season your campaign is running in. None of it is dangerous on its own, but it erodes the brand system you paid to build. For the deeper argument on why a codified brand system is the fix, see what is creative volume.
Half two: safety IN placement
The placement half has a familiar part and a new part.
The familiar part is adjacency: keeping your ads away from content that damages the brand. Platform brand-safety controls, inclusion and exclusion lists, and third-party verification are the usual tools for that job, and AI-made ads still need them.
The new part is the AI label. The rules come from three places:
- Platforms. TikTok's ad policy says "Significantly edited media and AIGC content is allowed if the following requirements are met: Apply the AIGC label, or by adding a clear disclaimer, caption, watermark, or sticker of your own," lists "Content that contains images, video, or audio that are completely AI-generated" among the content it covers, and says that if TikTok identifies undisclosed AI-generated content, "your ad will be rejected or restricted." Meta has required since January 11, 2024 that advertisers disclose when ads about social issues, elections or politics "use AI to create or edit photorealistic images, videos, or realistic audio" that depicts, for example, "A real person saying or doing something they didn't." Meta also applies "AI info" labels to ads made with its own generative AI tools, and in a June 1, 2026 update said it "will also begin automatically detecting ads created or edited using third-party AI tools through industry-standard signals." Google says "Advertisers must disclose all election ads that contain synthetic or digitally altered content" by ticking a checkbox in campaign settings; on some formats Google then generates the in-ad disclosure, and on the others the advertiser must provide "a prominent disclosure."
- US state law. New York's synthetic performer disclosure law for ads has been in effect since June 9, 2026, according to the Governor's office, and California's SB 1050 (chapter 246 of 2026) requires a "clear and conspicuous disclosure" when an ad prominently includes a synthetic performer, from January 1, 2027. Both laws are written for performers who are not recognizable as an identifiable real person.
- EU law. The European Commission says that under the AI Act "certain AI-generated content should be clearly and visibly labelled, namely deep fakes," and that the Act "became applicable on 2 August 2026, with some exceptions."
The conditions and exceptions for each rule are in do you have to disclose AI-generated ads. Treat the label as a compliance item on the upload checklist, not a performance variable to hide from.
Whether a label hurts performance is not settled, and we have no data of our own on it. One survey data point: in IAB's study, 73% of Gen Z and Millennial consumers said that if they knew an ad was created with AI, it would either increase or make no difference to their likelihood to purchase. The same study found that 89% of advertisers who have used generative AI to create ads at least sometimes disclose it, but less than half always do. If the label's effect matters to you, test it in your own account.
Sentiment is the other placement risk. IAB found that 82% of ad executives believe Gen Z and Millennial consumers feel very or somewhat positive about AI-generated ads, against 45% of consumers who actually do: a 37-point gap, up from 32 points in 2024. IAB's own conclusion is that "with clearer communication, an emphasis on high creative standards and more consistent disclosure practices, advertisers may be able to improve consumer attitudes toward AI-generated advertising." The defense is not to hide the AI. It is to make the craft good enough that "AI-made" and "well-made" stop being in tension.
The approval workflow blueprint
A QA gate only works if it is a fixed step, not a vibe. Here is a gate you can run before any AI-assisted asset reaches a client or an ad account. It is a final look, not a re-do, and it runs per deliverable set, not per asset.
Gate 1: Claims check. Every stated or implied claim maps to a substantiation source. No source, no claim. Owned by whoever owns legal or brand compliance. This is the gate that prevents the expensive mistakes.
Gate 2: Brand-mark integrity. Logos, wordmarks, product names, and packaging match the brand files exactly. Check at full resolution, not at thumbnail. Marks that matter get composited as overlays, not left to the model.
Gate 3: Likeness and rights. Any face or voice is either fully synthetic and not resembling a real person, or licensed with the paperwork on file. Backgrounds scanned for third-party trademarks. A synthetic person presents; it does not pose as a customer or describe an experience it never had.
Gate 4: Context and tone. The asset fits the brand system, the season, and the audience. No accidental cliches, no off-palette drift, no scene that reads wrong in the target market. For multi-market work, pair this with a localization pass.
Gate 5: Disclosure and placement. AI labels applied where the platform requires them (on TikTok ads, for completely AI-generated content as well), synthetic performer disclosures added where state law requires them, adjacency and exclusion lists set, upload checklist complete.
Budget fixed time per deliverable set for gates 1 through 4 (15 to 30 minutes is a sensible starting point to adjust), and fold gate 5 into the existing upload routine. Keeping the reference images, brand assets and every generation on one 8frame board is what makes this fast: reviewers compare the batch against the same source instead of chasing files. Inviting reviewers onto a board needs a Creator plan (collaboration up to 3 members) or an Agency plan (up to 30). See how the review step fits a full variant test in the AI ad variant testing workflow.
What this means for brand teams
Brand safety has mostly been a media-side function you bought from a verification vendor. In 2026 it is also a creative-side function you have to build, because a false claim or a drifted logo that shipped because nobody read the generated copy can embarrass a brand as much as a bad adjacency. The fix is not to slow generation down. It is to put a fixed, fast gate between generation and delivery, and to treat that gate as non-negotiable regardless of how the asset was made.
The forcing function is volume. If you are going to run many times the creative, you need a review step that scales with it, and "someone will notice" does not scale. A codified gate does. Build it once, run it on everything, and the speed advantage of AI stops being a liability.
FAQ
What is AI brand safety?
Two things. Safety of the creative: making sure AI-generated ads don't invent claims, distort logos, misuse likenesses or drift off-brand. Safety in placement: the familiar adjacency controls, plus the AI disclosure labels and consumer sentiment that now travel with AI-made ads.
Do I have to label AI-generated ads?
It depends on the platform, the ad and where it runs. TikTok's ad policy asks for a label on completely AI-generated content. Meta requires advertisers to disclose certain photorealistic AI content in ads about social issues, elections or politics, and Google requires disclosure of election ads with synthetic or digitally altered content. New York has required ads to disclose synthetic performers since June 9, 2026, California will from January 1, 2027, and in the EU the AI Act's labelling rule for deep fakes has applied since August 2, 2026. The details are in do you have to disclose AI-generated ads. This is general information, not legal advice.
Can a synthetic presenter endorse my product?
It can present the product and deliver claims that are true and substantiated. It should not pose as a customer or describe an experience it never had: the FTC's reviews and testimonials rule addresses testimonials by someone who does not exist or who had no actual experience with the product, and an endorsement "can't be used to make a claim the marketer of the product couldn't legally make." Check scripts that sit between the two with counsel.
How do consumers feel about AI-generated ads?
In IAB and Sonata Insights' January 2026 study, 45% of Gen Z and Millennial consumers felt positive about AI-generated ads, against 82% of ad executives who thought they would; the share feeling negative rose 12 points since 2024. Yet 73% said knowing an ad was made with AI would increase or not change their likelihood to buy.
Sources
- IAB and Sonata Insights, "The AI Ad Gap Widens", January 15, 2026 (survey of 505 US Gen Z and Millennial consumers and 104 US ad executives): iab.com, checked 2026-10-02
- FTC updated Endorsement Guides, June 29, 2023: FTC press release, checked 2026-10-02
- FTC, "FTC's Endorsement Guides: What People Are Asking": ftc.gov, checked 2026-10-02
- FTC Rule on the Use of Consumer Reviews and Testimonials: FTC announcement, August 14, 2024; effective date: Federal Register, August 22, 2024, checked 2026-10-02
- TikTok advertising policy, "Misleading and false content" (last updated April 2026): TikTok Business Help Center, checked 2026-10-02
- Meta, "Helping People Understand When AI Or Digital Methods Are Used In Political or Social Issue Ads" (November 8, 2023, updated February 19, 2026): Meta, checked 2026-10-02
- Meta, "Expanding GenAI Transparency for Meta's Ads Products" (updated June 1, 2026): about.fb.com, checked 2026-10-02
- Google Ads policies, "Political content", disclosure requirements for synthetic or digitally altered content: Google Ads policy help, checked 2026-10-02
- New York synthetic performer disclosure law (A.8887-B / S.8420-A): NY Assembly; Governor's announcement of June 9, 2026, checked 2026-10-02
- California SB 1050 (chapter 246 of 2026): bill text, bill history; effective date under the California Constitution, Article IV, Section 8: leginfo, checked 2026-10-02
- European Commission, "AI Act" (regulatory framework for AI): digital-strategy.ec.europa.eu, checked 2026-10-02
- 8frame collaboration limits by plan: 8frame pricing, checked 2026-10-02
Run your variants, brand assets, and review in one place. Build your creative on 8frame with image and video models on one canvas and your brand references on the same board, so the QA pass compares against a single source of truth. See the complete 2026 guide to AI in advertising for how safety fits the full stack.
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.