The European Commission has published a set of free icons that publishers, advertisers and platforms can use to mark AI-generated or manipulated content, a practical step toward compliance with transparency duties under the EU AI Act that begin applying on 2 August 2026.

The icons, released through the Commission's Shaping Europe's Digital Future portal with a last update dated 10 June 2026, form part of Section 2 of the Code of Practice on marking and labelling of AI-generated content. According to the Commission's published guidance, deployers of generative AI systems can use the icon set to label certain AI-generated content in accordance with the AI Act's transparency rules, and the icons support compliance with Article 50(4) of Regulation (EU) 2024/1689, which requires disclosure of deepfakes and of AI-generated text published on matters of public interest.

Three icons make up the set: a basic icon for general disclosure, a "Fully AI-Generated" icon for content created entirely by AI without human editorial control, and a "Partially AI-Modified" icon for pre-existing human-made content altered with AI. Each is available in four variations covering black, white, and 50 percent transparency versions, downloadable in SVG and PNG formats. According to the Commission, the icons underwent user-testing, and performance improved across all measures when the basic icon was accompanied by a text label such as "modified."

The publication follows months of parallel legislative activity. The AI Act itself, formally Regulation (EU) 2024/1689, entered into force on 1 August 2024, with its transparency provisions under Article 50 among the last major obligations to take effect under the original timetable. Those obligations require deployers of AI systems that generate deepfakes to disclose the artificial origin of that content, and require similar disclosure for AI-generated text published to inform the public on matters of public interest, unless a natural or legal person has exercised editorial control over the content.

What the icons cover, and what they do not

Not every piece of AI-generated content needs a label under the Commission's framework. According to the published guidance, the disclosure requirement covers only two categories: deep fakes, defined as AI-generated or manipulated image, audio, or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful; and AI-generated or manipulated text published to inform the public on matters of public interest, where that text did not undergo human review or editorial control and where no legal or natural person assumed editorial responsibility for it.

The Commission's guidance draws a further distinction for creative and satirical content. Where deep fake material forms part of an evidently artistic, creative, satirical, fictional or analogous work, the transparency obligation narrows to disclosure "in an appropriate manner that does not hamper the display or enjoyment of the work." That carve-out does not extend to advertising. A separate Bird & Bird analysis of the Commission's draft Article 50 guidelines, published 11 June 2026, states plainly that an "AI-manipulated video involving deep fake simulation of humans advertising a product in an AI-generated scene depicting the use of the product by the simulated consumers with the aim of persuading viewers to buy the product" does not qualify for the lighter artistic-content regime. The same draft guidelines exclude AI-generated imagery of celebrities implying involvement in activities they did not take part in, unless that imagery serves a genuine fictional, satirical or analogous purpose.

Two further exemptions apply. Deepfakes and public-interest text authorized by law for detecting, preventing, investigating or prosecuting criminal offenses fall outside scope. So does AI-generated text that underwent human review or editorial control, provided a named legal or natural person holds editorial responsibility for the publication.

The 200-token threshold and the marking-versus-labelling split

The icons address labelling duties under Article 50(4), which fall on deployers, the organizations that actually publish or distribute content to the public. A separate and technically distinct duty under Article 50(2) falls on providers, the companies that build the underlying generative AI systems, requiring them to mark outputs in machine-readable formats so the AI origin of content can be detected automatically.

According to a detailed analysis published 22 June 2026 by law firm Bird & Bird, the finalized Code of Practice on Transparency of AI-Generated Content dropped an option that had existed in earlier drafts: a "provenance certificate," or signed manifest, that providers could once use as an alternative to watermarking for text outputs. The final text now requires imperceptible watermarking for any free-form text longer than 200 tokens, with only very short text falling below that carved out. Providers must also maintain at least two machine-readable marking layers for most content: digitally signed and time-stamped metadata plus an imperceptible watermark, according to the Bird & Bird summary of Commitment 1 under Section 1 of the Code.

That marking-and-detection regime for providers is distinct from the icon-based labelling regime for deployers described in the Commission's newly published guidance, though the two work together. A generative AI system embeds a machine-readable mark; the organization publishing the resulting content then applies a human-visible icon, at least for deepfakes and qualifying public-interest text.

The Commission's guidance sets out where the icon must appear, and the rules are stricter than a passing mention deep in a document. According to the published summary, the icon must be clearly perceivable and distinguishable at the latest at the time of first exposure of a natural person to the content, must be placed where no intervening overlay elements exist, and must be directly embedded into the deep fake or published text, except for creative works, unless equivalent alternatives such as a user interface overlay are available. The icon must remain visible when content is reshared or downloaded.

A separate practitioner guide by product manager Gianna Brachetti-Truskawa, published 2 July 2026, states the placement requirement even more directly for text: the label "goes at the top, in the byline area or before the headline," and "a footer disclosure does not satisfy 'at first exposure.'" The same guide notes that the icon alone does not establish compliance with Article 50(4); deployers remain responsible for meeting the underlying legal requirement regardless of whether they use the Commission's icon or an equivalent of their own design.

The Commission's own guidance reinforces this point directly. The use of the EU icons is optional, but the labelling requirements under Article 50 of the AI Act are not, and the use of the icons does not by itself establish legal compliance. Signatories of the Code of Practice on marking and labelling of AI-generated content are required to duly implement the measures the Code contains.

An editorial-responsibility exemption, with a liability trade

Article 50(4) contains a narrower path for publishers who conduct genuine editorial review of AI-assisted text: content that underwent a process of human review or editorial control, where a natural or legal person holds editorial responsibility for the publication, falls outside the disclosure obligation entirely.

That exemption comes with a catch, according to the same practitioner guide by Brachetti-Truskawa. The exemption "removes the obligation but transfers liability." A decision-tree resource accompanying the guide frames the choice starkly: claiming editorial responsibility means an organization is "ready to own full liability if an error slips through," because "the AI hallucinated" stops being a valid defense once that responsibility is formally claimed. The alternative, applying the disclosure label instead, keeps liability attached to the AI system itself and does not require documentation of a review process. The guide recommends that organizations relying on the exemption document, before publication, who reviewed the content, when, and why they were qualified to do so, describing this as the organization's defense if a dispute arises later.

The distinction between a general editor and a reviewer with substantive, independently verifying expertise in the specific content domain matters under this framework. According to the guide, proofreading for style, without independently verifying whether a specific factual claim is accurate, does not meet the threshold; the exemption requires substantive review capable of catching factual errors and hallucinations in that particular subject area.

Research favors brief labels over lengthy disclosures

One data point in the practitioner guide is likely to influence how deployers actually word their labels. According to research the guide cites as published in 2026, long disclosures, including specific tool names and descriptions of oversight workflows, performed worse on reader trust than brief labels did. That finding aligns with the Commission's own icon design choice: only the capitalized acronym "AI" is mandatory, with additional detail such as "GENERATED" or "MODIFIED" merely encouraged rather than required, according to the Bird & Bird analysis of the finalized Code.

The Bird & Bird analysis notes a further reason the Commission may have kept the deeper disclosure optional rather than mandatory. Requiring every deployer to disclose whether content was "generated" versus merely "modified" could affect a work's copyright protectability, since the term "generated" implies potentially no human authorship, a status copyright law does not protect. Leaving that second-layer disclosure optional allows a deployer to describe content as "modified" where the facts support it, a characterization more favorable to copyright protectability and, according to the Commission's own user-testing, better understood by audiences in any case.

Organizations that sign the Code of Practice by 22 July 2026 gain what the practitioner guide describes as a presumption of compliance, shifting the focus of enforcement toward adherence to the Code's specific measures rather than toward the underlying legal text of Article 50 itself. Signatories are listed publicly. The Code itself states, according to the Bird & Bird analysis of its own text, that adherence "does not constitute conclusive evidence of compliance," meaning that even signing does not fully insulate an organization from scrutiny, though it does shift where that scrutiny concentrates.

For organizations that do not sign, the Code's influence remains indirect but real. Bird & Bird's analysis states that market surveillance authorities may treat the Code as the benchmark for what adequate compliance looks like, meaning a non-signatory company that adopts its own alternative technical approach carries the burden of demonstrating that its alternative is at least as effective, interoperable, robust, and reliable as the Code's approach.

A parallel deadline shift for high-risk systems does not touch this obligation

Article 50 transparency duties are proceeding on their original schedule even as other parts of the AI Act have shifted. In a provisional political agreement reached on 7 May 2026, the Council of the EU and the European Parliament settled on new fixed dates for a separate category of obligations, those covering high-risk AI systems under Chapter III of the Act, such as systems used in employment screening, credit scoring, and biometric identification, moving compliance for Annex III systems to 2 December 2027 and for Annex I systems to 2 August 2028. That agreement forms part of the Digital Omnibus package the European Commission introduced on 19 November 2025 and does not touch the Article 50(4) labelling duty or the 2 August 2026 date on which it takes effect. A separate related change under that same agreement gives providers of generative AI systems already on the market before 2 August 2026 until 2 December 2026 to bring their machine-readable marking under Article 50(2) into conformity, according to reporting on the Digital Omnibus agreement.

Why this matters for advertising and marketing organizations

The icon publication lands squarely inside a compliance question that advertising agencies, in-house marketing teams, and publishers using generative AI for campaign assets have had reason to track since the Commission opened its Article 50 consultation in September 2025. The distinction the Bird & Bird analysis draws between advertising content and genuinely artistic or satirical work removes any expectation that a persuasive product demonstration featuring AI-generated consumer avatars could rely on the lighter creative-content disclosure regime; ordinary full labelling applies.

The editorial-responsibility exemption is likely to matter most for organizations publishing AI-assisted commentary, market analysis, or news content on matters of public interest, categories the Commission's guidance defines broadly enough to include current affairs, public health, environmental issues, politics, consumer protection, economic policy and science, according to a decision-tree resource accompanying the practitioner guide. Pure product marketing, according to the same resource, sits outside the Article 50(4) public-interest scope entirely, though other Article 50 obligations may still apply depending on how the content is generated and disseminated.

For marketing organizations relying on multiple vendors, freelance contributors, or agency partners to produce AI-assisted content, the uniformity question the Bird & Bird analysis raises is a practical one: several parties feeding one brand may each apply their own compliant but differently designed label, leaving a single advertiser with a patchwork of inconsistent marks across its content. Standardizing on the Commission's EU icon rather than a bespoke equivalent, and building contractual requirements around the Code's placement rules, is one way organizations are addressing that risk, according to the analysis.

Timeline

  • 13 June 2024 -- The European Parliament and Council adopt Regulation (EU) 2024/1689, the EU AI Act.
  • 1 August 2024 -- The AI Act enters into force.
  • 4 September 2025 -- The European Commission opens a consultation on guidelines and a Code of Practice for Article 50 transparency obligations.
  • 8 May 2026 -- The Commission publishes its draft Article 50 guidelines, opening a stakeholder consultation period running until 3 June 2026.
  • 7 May 2026 -- The Council and European Parliament reach a provisional agreement under the Digital Omnibus package, fixing new 2027 and 2028 deadlines for high-risk AI systems without altering the Article 50(4) labelling deadline.
  • 10 June 2026 -- The European Commission publishes the free EU icons for labelling AI-generated content, last updated on this date.
  • 11 June 2026 -- Bird & Bird publishes an analysis of the Commission's draft Article 50 deep fake guidelines.
  • 22 June 2026 -- Bird & Bird publishes its analysis of the finalized Transparency Code of Practice.
  • 2 July 2026 -- A practitioner decision-tree guide on the Article 50(4) disclosure-versus-editorial-responsibility choice is published.
  • 22 July 2026 -- Deadline for organizations to sign the Code of Practice to obtain a presumption of compliance.
  • 2 August 2026 -- Article 50 transparency obligations, including the deepfake and AI-text labelling duty, become legally applicable.
  • 2 December 2026 -- Deadline for generative AI providers already on the market before 2 August 2026 to bring Article 50(2) machine-readable marking into conformity.
  • 2 February 2027 -- Deadline for providers to implement a watermark-detection interoperability solution.

Summary

Who: The European Commission published the icons; they apply to deployers of generative AI systems, including advertisers, marketing agencies, publishers and platforms operating in or serving the European Union.

What: A free set of three icons, each with four visual variations in SVG and PNG format, that organizations can use to label deepfakes and AI-generated text under Article 50(4) of the EU AI Act. The icons form part of the Code of Practice on marking and labelling of AI-generated content and were published alongside detailed placement rules distinguishing embedded labels from overlay alternatives.

When: The icons carry a last-update date of 10 June 2026. The underlying legal obligation they support, Article 50 of Regulation (EU) 2024/1689, becomes applicable on 2 August 2026. Organizations seeking a presumption of compliance through the Code of Practice must sign by 22 July 2026.

Where: The obligation applies across the European Union and European Economic Area, and extends to organizations established outside the EU if their AI-generated content is made accessible to readers within the bloc.

Why: The labelling requirement addresses the risk that AI-generated or manipulated content, particularly deepfakes and AI-written material on matters of public interest, could deceive audiences about its authentic origin. Advertising content is explicitly excluded from the lighter disclosure regime available to genuinely artistic or satirical work, meaning marketing organizations using AI-generated consumer avatars or synthetic spokespeople in persuasive campaigns face the full labelling obligation. Non-compliance with Article 50 falls under the AI Act's general penalty structure, which permits fines of up to 15 million euros or 3 percent of worldwide annual turnover, whichever is higher, for violations other than the more severely penalized prohibited practices under Article 5.