TL;DR
- The IAB released Part 2 of its AI Transparency and Disclosure Framework on 18 August 2026, its first update since January 2026.
- Its central recommendation is targeted disclosure rather than universal labeling, particularly when AI use materially affects authenticity, identity, or representation in ways that could mislead consumers.
- AI-prompted images and video, some synthetic voices, avatars, digital twins and AI chatbots should be disclosed; routine post-production, background music, stylized avatars and text or copy need no disclosure.
- Caroline Giegerich, the IAB's vp of AI, said the update was driven by legislation maturing in New York, California, South Korea and the EU, and conceded the legislative section would date almost immediately.
The Interactive Advertising Bureau released Part 2 of its AI Transparency and Disclosure Framework on 18 August 2026, its first update since the original guidance published in January 2026. The framework's central recommendation is "targeted disclosure rather than universal labeling, particularly when AI use materially affects authenticity, identity, or representation in ways that could mislead consumers".
Caroline Giegerich, the IAB's vp of AI, told Marketing Brew the update was driven by AI-disclosure legislation maturing in New York, California, South Korea and the European Union. Two framing points belong up front, because the document is easy to over-read.
This is not the IAB's AI visibility measurement framework
The IAB has published more than one AI framework in 2026, and they are not the same document. Earlier in August 2026 the IAB released a framework for measuring AI visibility, which is about how brands appear inside AI systems. The 18 August 2026 release is a separate publication on a separate subject: whether and how advertisers should disclose that AI was used to make a piece of creative. Anyone searching for one and landing on the other will draw the wrong conclusion.
The IAB AI Transparency and Disclosure Framework is voluntary
State this plainly, because it changes what a marketing team should do with it. The IAB AI Transparency and Disclosure Framework published on 18 August 2026 is a voluntary industry framework. It is not a law, and it is not an ad platform's policy. Ad platforms set their own disclosure requirements inside their own advertising policies and enforce them by rejecting or labelling creative; legislatures pass statutes and attach penalties; an industry body publishes guidance and hopes members adopt it. This is the third kind. Giegerich argued the framework functions as "the voice of the industry", which is an accurate description of its authority and also its limit.
What the IAB says to disclose, and what it says needs no disclosure
The two lists are the substance of the 18 August 2026 update. The IAB suggests a standardized sparkle icon or a text label as the disclosure mechanism where disclosure applies. The table reproduces both lists as published.
| AI use in advertising | IAB Part 2 position, 18 August 2026 |
|---|---|
| AI-prompted images and video | Disclose |
| Some synthetic voices, avatars, and digital twins of deceased or living individuals | Disclose |
| AI chatbots or assistants | Disclose |
| Routine post-production, internal workflows, clearly fantastical imagery, background music, standard audio enhancement | No disclosure |
| Authorized synthetic voices of real individuals, generic synthetic voices, digital twins in standard brand endorsement, obvious cartoon or stylized avatars, text or copy | No disclosure |
The copy carve-out is the line that matters most to marketers
In the framework the IAB published on 18 August 2026, text or copy sits on the no-disclosure list. AI-written ad text, under this framework, does not require a label.
Analysis: that is the most consequential single item in the document, given how much AI-assisted ad copy is already running. It also follows the logic of the rest of the framework rather than contradicting it. The test the IAB set is whether AI use materially affects authenticity, identity or representation. A headline generated by a model still makes a claim in the advertiser's own voice, and the advertiser remains answerable for whether the claim is true. Nobody's face, voice or likeness is being synthesised. The identity in the ad is still the brand's.
The practical consequence is that a team should stop treating "we used AI" as the trigger question. Under this framework the trigger is narrower: did AI produce something a consumer would take as a real person, a real performance or a real result?
Applying the material-effect test to actual ad assets
This section is analysis, not IAB text. The test the IAB published on 18 August 2026 is readable but it has to be applied asset by asset, and the boundary is not always obvious.
- A product photo where AI removed a stray cable is routine post-production under the IAB's own list. The same photo where AI generated the product's finish is a representation question.
- A voiceover using a generic synthetic voice sits on the no-disclosure list. The same script read by a synthetic version of a named presenter is on the disclose list unless the use is authorized, and authorization is doing heavy work in that sentence.
- An obviously stylised cartoon avatar needs no disclosure. A photoreal avatar that a viewer would read as a person is a different asset even if the production process was identical.
- A before-and-after image is the hardest case, because the AI question and the claims question are the same question.
The under-eye cream example is a claims problem as much as an AI problem
Giegerich framed the test as consumer deception, and her example to Marketing Brew was a shopper who buys an under-eye cream after a TikTok endorsement and then discovers the depicted results were AI-generated.
Analysis: that example describes two failures stacked on each other. One is a disclosure failure, which is what the framework addresses. The other is a claims-substantiation failure, which existed long before generative tools and is already regulated in most markets under advertising and consumer protection rules. If a depicted result is not achievable, showing it is a problem whether a model rendered it, a retoucher rendered it, or a studio lit it that way. The AI framing makes the failure easier to spot; it did not create it. A team that only asks "was this AI-generated" and never asks "can we substantiate this result" has solved the smaller half.
Giegerich conceded the legislative section will date quickly
The IAB's vp of AI told Marketing Brew on 18 August 2026 that the framework's legislative section would be out of date almost immediately, while arguing the document still functions as the voice of the industry. That is an unusual admission to publish alongside a framework and it is worth taking at face value. New York State's law requiring transparency about "synthetic performers" in ad materials took effect earlier in 2026, and Giegerich named New York, California, South Korea and the European Union as the jurisdictions where disclosure legislation is maturing.
The timeline the source establishes, in three dated points:
| Date | What happened |
|---|---|
| January 2026 | IAB published the original AI Transparency and Disclosure Framework guidance |
| Earlier in 2026 | New York State law on transparency about "synthetic performers" in ad materials took effect |
| 18 August 2026 | IAB released Part 2, the first update to the framework |
What the source did not say
- It described no enforcement mechanism. The IAB does not police adoption and no penalty was mentioned.
- It gave no compliance deadline. There is no date by which an advertiser is expected to adopt the framework.
- It did not address Thailand or any Asian jurisdiction other than South Korea, which was named only as a place where legislation is maturing.
- It did not specify the design, placement, size or duration of the sparkle icon or text label beyond suggesting them as the mechanism.
- It did not state that any ad platform has adopted the framework, and adoption by a platform would be a separate event.
What this means for Thai marketers
Thailand has no AI-labelling law for advertising, and the 18 August 2026 IAB framework does not create one anywhere. What it creates is a defensible default for teams that currently have none.
The exposure that is real rather than theoretical is jurisdictional. Thai brands and the agencies producing for them are in scope the moment creative runs into the European Union, California, New York or South Korea, which Giegerich named as the jurisdictions driving this update. A Bangkok team producing a campaign that will also run in a European market is producing under someone else's rules regardless of what Thai rules say. Beyond the fact that legislation is maturing in those four places and that the New York synthetic performers law took effect earlier in 2026, the source did not describe the content of any specific statute, and this post is not going to fill that gap with guesses.
The domestic read is different. Local production of AI-generated KOL-style visuals, synthetic voices and AI avatars is already routine in Thai campaigns, and the material-effect test is available to adopt now without waiting for a Thai rule. The cheap version is a two-column note in the creative brief: which AI uses on this asset touch identity, likeness or a claimed result, and which are production work. That question belongs in the same conversation as the rest of a brand's content production process, and it lands hardest on social creative, where synthetic presenters and endorsement-style formats are most common.
FAQ: the IAB AI disclosure framework
Is the IAB framework a law I have to follow?
No. It is a voluntary industry framework published by the Interactive Advertising Bureau on 18 August 2026, with no enforcement mechanism and no compliance deadline described in the source, and it is separate from both statute and any ad platform's own policy.
Do I have to label ad copy written by AI?
Not under this framework. The IAB places text or copy on its no-disclosure list, alongside routine post-production, internal workflows and background music, so AI-assisted ad text does not carry a label under the 18 August 2026 guidance.
Does this apply in Thailand?
The source did not address Thailand. Thailand has no AI-labelling law for advertising, and the IAB framework is voluntary everywhere, but creative running into the European Union, California, New York or South Korea falls under whatever those jurisdictions require.
Is this the same as the IAB AI visibility framework from earlier this month?
No, they are different documents on different subjects. The earlier IAB release covered measuring AI visibility; the 18 August 2026 release covers disclosing AI use in creative.
What actually triggers a disclosure?
The IAB's test is whether AI use materially affects authenticity, identity or representation in ways that could mislead consumers, which in practice covers AI-prompted images and video, some synthetic voices, avatars, digital twins of deceased or living individuals, and AI chatbots or assistants.
If a brand is producing AI-assisted creative and wants a defensible position written down before a platform or a regulator asks for one, Relevant Audience works on that alongside AI search and GEO.
