The AI Ad Disclosure Requirement Is Here, From Law and From the Industry Itself

Starting in July 2026, Google will let advertisers add AI disclosure labels directly onto image and video ads without running afoul of its rules against text overlays and watermarks. Google Ads will also be able to apply these labels automatically to assets built with its own AI tools. This is not Google being generous. It is Google catching up to a requirement that is closing in from every direction: law in multiple jurisdictions and the ad industry's own standards bodies.

The Legal Requirement

This is no longer a single law in a single place. At least four jurisdictions now require disclosure when AI generated or AI edited content shows up in an ad.

• European Union: The AI Act's Article 50 requires providers of generative AI systems to mark synthetic audio, image, video, and text output in a machine readable format, effective August 2, 2026.
• India: The Ministry of Electronics and IT's 2026 amendment to the intermediary rules, paired with the Advertising Standards Council of India's draft labelling guidelines, requires prominent disclosure and embedded provenance metadata for AI generated ad content that could sway a consumer's decision.
• New York: Its synthetic performer disclosure law, in effect since June 2026, requires ads containing AI generated human performers to disclose it, with penalties running up to $5,000 per repeat violation.
• California: The AI Transparency Act, as amended by AB 853, requires covered generative AI providers to attach both a visible label and hidden metadata to AI generated image, video, and audio content, enforced by the state Attorney General at $5,000 per violation.

None of these are suggestions advertisers can opt into. Google says plainly that its new label setting "doesn't guarantee compliance with specific regulations," and advertisers are told to seek their own legal guidance for whatever jurisdiction their ads reach.

Industry Standards Are Catching Up Too

Law is not the only pressure here. In January 2026, the IAB Tech Lab released its AI Transparency and Disclosure Framework, the industry's first, and it takes a risk based approach: any ad that could plausibly deceive a viewer, an AI generated person in a fabricated before and after shot, for example, must be disclosed. The framework recognizes two ways to do that. One is a consumer visible cue: a label, badge, or icon placed on or next to the creative. The other is a backend, machine readable disclosure built on C2PA protocols, the same open standard supported by Adobe, Google, Microsoft, Meta, and the BBC.

The 4A's is pushing in the same direction from the agency side. Its content provenance work calls for "uniform, technical standards that certify the origin and lineage of media content" and for labeling protocols consistent enough to work the same way across every platform and jurisdiction. Like the IAB, the 4A's draws a line at deception: it does not treat every AI assisted edit, translation or image enhancement among them, as something that needs a disclosure. What it wants flagged, in a standardized way, is AI use that could mislead. Two industry bodies, working from different sides of the ad business, have landed on the same answer: certify the origin of the content itself, do not just caption it.

Where A Pasted On Label Runs Out

Google's new setting addresses the first half of that answer. It gives advertisers a supported way to add a visible label to a creative. It does not address the second half. A text label sitting on top of an image is just more pixels. It can be cropped, blurred, screenshotted, or removed once the asset leaves the ad platform and starts moving across the internet, and nothing about it survives re encoding or re upload in a way anyone can check. That is exactly why Google's own guidance stops short of calling it compliant. A visible label tells a viewer what to believe. It does not let anyone verify the claim.

What Limbo Builds

C2PA closes that gap by embedding cryptographically signed metadata inside the file itself: who created it, what tools touched it, and whether AI was involved at any stage. That record travels with the asset through cropping, compression, and platform transfers, and it can be checked independently rather than taken on faith. As the EU, India, New York, California, the IAB, and the 4A's all converge on the same conclusion, that disclosure has to be machine readable and verifiable, advertisers need infrastructure for the backend half of this requirement, not just a checkbox for the visible half.

Limbo builds that infrastructure. If your ad content needs to prove what it is, not just claim what it is, get in touch.