EU AI Act Article 50 Transparency Code of Practice June 2026: What AI Ad Creative Must Disclose Before August 2

What Changed in June 2026

On June 10, 2026, the European Commission published a voluntary Code of Practice on the transparency of AI-generated content, the latest instrument supporting the EU AI Act's Article 50 transparency obligations. It follows the Commission's draft Guidelines on the implementation of the Article 50 transparency obligations, published on May 8, 2026 and open for stakeholder consultation until June 3, 2026.

The timing matters for advertisers. Article 50 — the AI Act provision governing transparency for AI systems that interact with people or generate synthetic media — becomes applicable on August 2, 2026. The Code of Practice and the Guidelines are the first concrete interpretive support the Commission has offered across the full scope of Article 50, and they arrive with under two months on the clock.

For any brand or agency running AI-generated images, video, voice, or text in EU-facing campaigns, this is the moment to convert a future obligation into a live workflow. The obligations attach to both providers of generative AI systems and to the deployers who use those systems to produce advertising creative.

"Article 50 is not a creative-freedom question; it is a disclosure-and-marking question. If an EU user could mistake AI-generated ad content for authentic human-made or real-world content, the AI Act expects it to be labeled and machine-readable by August 2, 2026.
— AuditSocials analysis of the European Commission's Article 50 Guidelines"

This guide explains the difference between the voluntary Code, the non-binding Guidelines, and the binding law; the four transparency obligations; what specifically changes for AI ad creative; and the steps to be ready. Track ongoing EU enforcement on the Policy Change Tracker and review the broader EU framework in the European Union compliance reference.

Code of Practice vs Draft Guidelines vs the Law

Three distinct instruments are in play, and conflating them is the most common mistake. The binding obligation is the AI Act itself; the Guidelines interpret it; the Code of Practice offers a practical, voluntary route to demonstrate alignment.

Voluntary does not mean optional in practice. Signing or following a Code of Practice is one of the clearest ways to evidence that an organization took reasonable steps toward compliance, which is relevant when a regulator assesses good faith. But the Code is a means, not the obligation — the obligation is the law, and it applies whether or not a company adopts the Code.

Because the Guidelines remain in draft and the Code is voluntary, advertisers should treat the specific mechanics as evolving and confirm the final text against the Commission's official publications rather than relying on summaries. The direction, however, is settled: transparency for AI-generated content is mandatory from August 2, 2026.

Article 50: The Four Transparency Obligations

Article 50 sets out transparency duties split between providers (who build and supply AI systems) and deployers (who use them). Four obligations are relevant to advertising and marketing operations.

For most advertisers the load-bearing duties are 50(2) and 50(4). If a campaign uses a generative system to create or materially alter creative, the underlying system should mark the output in a machine-readable way, and the deployer should ensure deepfake-style content is disclosed to viewers in a clear and distinguishable manner. The machine-readable mark and the human-facing label are complementary, not interchangeable.

What This Means for AI Ad Creative

The practical effect is that AI-generated advertising in the EU now needs two layers of transparency: a technical mark embedded in the asset, and a disclosure the viewer can actually perceive. Neither alone is sufficient if the content would otherwise mislead.

Generative tools increasingly emit machine-readable provenance signals such as C2PA content credentials, and several major model providers watermark their outputs. Deployers should not assume those marks are present by default — confirm with each tool whether its output carries a machine-readable AI mark, and preserve that mark through your editing and export pipeline rather than stripping it during post-production.

On the human-facing side, the disclosure has to be clear and appropriately placed: an on-frame label for video, a visible caption-level disclosure for image posts, and a perceivable cue for synthetic audio. The standard is whether an ordinary EU viewer would understand that the content is AI-generated before being misled by it. Platform-native AI labels (such as those offered inside Meta, TikTok, and other ad systems) help, but advertisers should not assume a platform tag alone discharges the Article 50 deployer duty — coordinate the platform label with your own in-creative disclosure. Run creative through the AI Compliance Audit to flag undisclosed synthetic elements before submission.

This obligation sits alongside, not instead of, platform AI-content rules and existing misleading-advertising law. A synthetic endorsement that is properly AI-labeled can still violate consumer-protection rules if it implies a real person endorsed a product they never used. Treat Article 50 as the transparency floor, not the whole compliance picture.

A second practical wrinkle is durability of the machine-readable mark across the ad-delivery chain. Creative typically passes through multiple hands — a generative tool, an editing suite, an asset-management system, and finally a platform's ad pipeline — and each handoff is an opportunity for provenance metadata to be stripped or lost. A mark that exists when the asset leaves the generative tool but is gone by the time the ad serves does not achieve the transparency the obligation is aiming for. Advertisers should therefore test the full pipeline, not just the source tool, and confirm the mark survives to the published asset.

The split of duties between providers and deployers also means accountability can be shared rather than singular. The provider of the generative system carries the marking duty for its outputs; the deployer using that output in a campaign carries the disclosure duty to the audience. In an agency-brand-platform chain, it is worth settling in advance who is responsible for which layer, because a gap between provider marking and deployer disclosure is exactly where an undisclosed-synthetic-content problem tends to appear.

Pre-August 2 Compliance Steps

The window between now and August 2, 2026 is short. A defensible readiness program covers tooling, workflow, and documentation.

• Inventory AI in the creative pipeline: map every tool that generates or materially alters ad images, video, voice, or copy, and record which ones emit a machine-readable AI mark.
• Preserve provenance end to end: ensure editing, resizing, and export steps do not strip C2PA or watermark metadata before the asset is published.
• Standardize human-facing disclosure: adopt a consistent on-creative label pattern for AI-generated video, image, and audio across EU campaigns.
• Coordinate with platform labels: enable platform AI-content disclosure where available and align it with your in-creative label rather than relying on one alone.
• Document decisions: keep a per-campaign record of what was AI-generated, what mark/label was applied, and why — the same evidence that supports a good-faith assessment under the Guidelines.
• Watch the final text: the Guidelines are still draft and the Code is new; confirm specifics against the Commission's official publications before locking processes.

For cross-jurisdiction exposure beyond the EU — for example synthetic-media and disclosure rules emerging in the US and UK — map obligations with the Legal Compliance Scan so a single AI creative does not pass in one market and fail in another.

Penalties and Enforcement

The AI Act's penalty structure is tiered by the type of breach. For breaches of obligations other than the most serious prohibited-practices tier, the Act provides for fines of up to EUR 15 million or 3% of total worldwide annual turnover, whichever is higher, with proportionality for the specific obligation and actor. Transparency duties under Article 50 fall into this general-obligations space rather than the prohibited-practices tier.

Enforcement is administered through national market-surveillance authorities designated by each Member State, coordinated at EU level. Because Article 50 becomes applicable August 2, 2026, the practical enforcement posture in the months after will set expectations — but advertisers should not plan around a grace period that has not been promised. The defensible assumption is that the obligation is live on August 2.

Enforcement also will not stand alone. Article 50 transparency sits within a wider EU framework that already governs advertising — the Digital Services Act's rules on advertising transparency, consumer-protection law on misleading commercial practices, and, for AI used in sensitive contexts, other parts of the AI Act. An advertiser that treats AI labeling as a single checkbox risks missing that the same synthetic creative can be assessed under more than one regime at once. The efficient posture is to build AI transparency into the creative workflow as a default and to coordinate it with the broader compliance review rather than running it as an isolated, last-minute step before a campaign goes live.

For a brand, the reputational exposure can outweigh the headline fine: AI-generated advertising that is later revealed as undisclosed synthetic content invites both regulatory scrutiny and public-trust damage. The transparency the AI Act requires is, in practice, also the transparency that protects the brand.

The strategic way to frame this internally is that Article 50 is converging with where audiences and platforms are already heading. Viewers increasingly expect to know when content is synthetic, platforms are rolling out their own AI-content labels, and provenance standards are becoming part of the creative supply chain regardless of regulation. An advertiser that builds AI transparency now is not only meeting an August 2026 legal obligation; it is getting ahead of an industry norm that would arrive even without the AI Act. The teams that treat the deadline as a forcing function to standardize disclosure and provenance across all markets — not just the EU — will spend less in aggregate than those that build a narrow EU-only patch and then retrofit again when the next jurisdiction follows. The cheapest compliance is the one designed once, applied everywhere, and maintained as a default rather than rediscovered per campaign.

AI Creative Transparency Checklist

• [ ] Every AI tool in the ad pipeline inventoried, with machine-readable-mark status recorded
• [ ] C2PA / watermark provenance preserved through editing and export
• [ ] Standard on-creative AI disclosure applied to EU video, image, and audio
• [ ] Platform-native AI label enabled and aligned with in-creative disclosure
• [ ] Deepfake / synthetic-spokesperson content clearly disclosed to viewers
• [ ] Emotion-recognition or biometric-categorisation tooling disclosed where used
• [ ] Per-campaign documentation of AI use, marks, and labels retained
• [ ] Final Guidelines / Code text confirmed against official Commission publications
• [ ] Cross-market synthetic-media obligations (US/UK) checked for the same creative

For the underlying obligation breakdown and the broader marketer view, see our companion guide on EU AI Act Article 50 advertising compliance, and keep enforcement developments tracked on the Policy Change Tracker.