New AI Labeling Requirement in August 2026: A Guide to Labeling AI-Generated Content
Today, artificial intelligence (AI) generates images, videos, text, and voices that are virtually indistinguishable from real recordings. What companies see as a gain in efficiency is increasingly becoming a problem for consumers: Where does the real recording end, and where does the synthetic creation begin? This is precisely where the EU’s new AI labeling requirement comes into play. Under Article 50(4) of theEU AI Act, mandatory transparency requirements for AI-generated content will take effect on August 2, 2026. This affects not only large technology providers but virtually every company that uses generative AI in a professional context. Violations are subject to fines of up to 15 million euros or three percent of global annual revenue.
Executive Summary – AI Disclosure Requirements at a Glance
- Effective Date: August 2, 2026: As of this date, the transparency requirements set forth in Article 50 of the EU AI Regulation will take effect—with substantial fines for violations.
- Operators are affected: Every company, agency, association, and government agency that uses generative AI for professional purposes must comply with the labeling requirements—not just the AI providers themselves.
- No blanket requirement: The regulation distinguishes based on content type, purpose, and individual. Purely private use is exempt, whereas commercial contexts are almost universally covered—in particular, deepfakes in images, videos, and audio, as well as texts on topics of public interest that are not subject to editorial control.
- The label must be visible: Hidden information in the legal notice, alt text, or a tiny watermark is not sufficient—the label must be clear, concise, directly associated with the content, and accessible to all users.
- Double Risk: In addition to fines under the AI Act, companies may face cease-and-desist letters under competition law pursuant to the UWG—and the reverse practice of “AI washing” (giving the false impression of using AI) can also lead to legal problems.
What is the AI labeling requirement, and where does it come from?
The AI labeling requirement is a central component of European AI legislation, better known as the EU AI Act. It sets forth the conditions under which content—including images, videos, audio, and text—generated or manipulated using artificial intelligence must be disclosed as such to the public. The legal basis is Article 50(4) of the AI Regulation. The provision is divided into two parts:
Image, Audio, and Video Content: Operators of an AI system that generates or manipulates image, audio, or video content constituting a deepfake must disclose that the content has been artificially generated or manipulated.
Text Content: Operators of an AI system that generates or manipulates text intended for publication to inform the public about matters of public interest must also disclose that the text was artificially generated or manipulated —unless the text has undergone human review or editorial oversight and a responsible person is accountable for its publication. Additionally, Article 50(5) of the AI Regulation applies: The disclosure must be made no later than the first interaction or perception—and must be clear, unambiguous, and accessible.
The goal of the AI labeling requirement is clear: to protect the public from disinformation, strengthen trust in digital content, and safeguard the democratic formation of public opinion in an era when synthetic content can be generated en masse and in real time.
Who is affected by the AI labeling requirement, and who is required to label AI-generated content?
The AI labeling requirement does not apply to the developers of AI systems, but rather to their operators—that is, anyone who uses AI tools under their own responsibility. According to Article 3(4) of the AI Regulation, an operator is any natural or legal person, public authority, agency, or other body that uses an AI system under its own responsibility—except in the case of purely personal, non-professional activities.
Affected are:
- Companies of all sizes that use AI tools in their business operations
- Advertising agencies, marketing and creative departments
- Editorial Offices and Publishing Houses
- Government agencies, public institutions, and research institutes
- Self-Employed Individuals, Freelancers, and Solopreneurs
- Influencers and content creators for commercial use
- Small shops that regularly sell on platforms such as eBay or Vinted and use AI-generated images as product photos
- Clubs, as soon as they appear in public, name sponsors, or charge admission
The following is not affected:
- Strictly personal use—such as sharing an AI-generated image of your uncle as an astronaut in a family chat
- Private conversations in closed messenger groups
- Personal posts not related to business
What AI-generated content must be labeled under the AI labeling requirement?
AI-Generated Images, Videos, and Audio: The Term "Deepfake"
The key factor triggering the labeling requirement for visual and audio media is the concept of “deepfakes.” Unlike in everyday language, this term is defined very broadly in legal terms. According to Article 3(60) of the AI Regulation, a deepfake is any image, audio, or video content generated or manipulated by AI that resembles real persons, objects, places, institutions, or events and would falsely appear to a person to be genuine or truthful.
Two criteria must be met:
- Realism: The content depicts something that actually exists or could plausibly exist.
- Potential for deception: The content appears so realistic that the average viewer might mistake it for the real thing.
Important: What matters is not the intent to deceive, but the effect on the audience. Even a completely fictional but photorealistically rendered person is a deepfake as defined by the regulation.
Common types of deepfakes in a business context:
- Photorealistic lifestyle images featuring AI-generated but lifelike people
- Virtual influencers and AI avatars that simulate human product reviews
- Voiceovers and cloned voices in commercials, podcasts, or explainer videos
- Rendering of product prototypes in real-world environments that looks like actual photography
- Edited real photos where the AI alters the content (not: simple color correction or the removal of minor image artifacts)
The following are generally not subject to labeling requirements:
- Illustrations, comics, or cartoon styles that are obviously artificial
- Stylized drawings or abstract key visuals
- Fantasy motifs that clearly cannot be real (e.g., a dragon above Cologne Cathedral)
- Standard image editing, such as retouching, color correction, or cropping
AI-Generated Texts: The "Public Interest" Filter
Stricter rules apply to text. Only AI-generated or manipulated text that is published to inform the public about matters of public interest—and that has not undergone any human editorial review—is subject to the labeling requirement.
Matters of public interest typically include:
- Up-to-the-minute news and political coverage
- Articles on Elections, Social Crises, and Public Safety
- Medical Information and Health Alerts
- Official Notices and Warnings
- Articles on Climate, the Environment, Consumer Protection, and Infrastructure
The following are generally not included:
- Traditional E-Commerce Product Descriptions
- Purely promotional copy and landing pages
- “About Us” texts with no social relevance
- Internal emails, memos, and internal company communications
The editorial exception (human-in-the-loop):
Even for texts of public interest, the labeling requirement does not apply if the text has been reviewed for content, edited, and approved by a human. Simply skimming the text for spelling errors is not enough—a substantive review of the content with a clear assignment of responsibility is required. In most companies, this exception applies provided that AI-generated texts are reviewed and approved by qualified personnel before publication.
Chatbots and Voicebots
Systems used in customer interactions are subject to an additional transparency requirement: People must be able to recognize that they are communicating with a machine. A voicebot on a telephone hotline should clearly state at the beginning of the call: “You are speaking with an AI assistant.” A clear indication is also required for text chatbots.
Guide: How to Implement the AI Disclosure Requirement in Practice
Article 50(5) of the AI Regulation provides clear guidelines on the format of the label. Two requirements are central:
The label must be visible no later than the time of the first interaction or notice. A notice that the user can only find by actively searching for it is not sufficient.
In practical terms, this means:
- Images: The disclosure must be clearly visible on or immediately next to the image—for example, at the bottom of the image, as a watermark, or directly in the caption. A hidden disclosure in the HTML alt text is not sufficient.
- Videos: The label should appear at least at the beginning and—for longer videos—at regular intervals. It is also recommended to include a note in the description so that the label is recognizable even when the video is played on mute.
- Audio Files and Podcasts: A spoken note at the beginning of the recording, supplemented by a visible note in the title, episode description, or metadata.
- Chatbots and voicebots: A clear message at the start of the interaction—for example: “Hello, you’re speaking with an AI assistant.”
The labeling must be clear, unambiguous, easily visible, and accessible to all. A tiny gray note at the edge of an image or an inconspicuous hashtag in the legal notice is not sufficient.
Examples of appropriate phrasing include:
- “Image created with AI”
- “Image generated by AI”
- “Video generated by AI”
- “Audio Created with AI”
- “Artificially generated voice”
- “Created with AI” or “Manipulated with AI”
The European Commission also provides symbols and a voluntary code of conduct that companies can use as a guide. It recommends the use of visible icons or labels with the abbreviations “AI” or “KI,” combined with brief text descriptions.
The following are insufficient:
- General notices in the legal notice or footer
- Barely visible watermarks without accompanying text
- Unobtrusive hashtags hidden within a longer post
Notes that become visible only after clicking or scrolling
What are the risks of violating the AI labeling requirement?
The AI labeling requirement is not a non-binding set of guidelines, but rather binding EU law with tangible consequences.
Fines Under the AI Act
Competition Law Risks (UWG)
"AI Washing" as a Reverse Risk
Reputational Risks
Additional Legal Risks
Our consulting services related to AI labeling requirements and AI governance and compliance
Ventum Consulting supports companies in implementing AI labeling requirements—as part of a comprehensive AI governance framework. The following components of our service portfolio are particularly relevant to this topic:
AI Governance & Compliance
We establish the governance structures, guidelines, and audit trails required by the EU AI Act and the GDPR. Specifically, with regard to the AI labeling requirement, this means clear guidelines, roles, and approval processes for handling AI-generated content—all of which are auditable and practical.
Risk Management for AI Content
We identify, assess, and address legal, regulatory, and antitrust risks associated with the use of generative AI—from deepfake risks and pitfalls under the Unfair Competition Act to copyright and personality rights.
AI enablement & training
We train marketing, editorial, communications, and legal teams on AI disclosure requirements—in a concise, practical manner, and with ready-to-use templates for disclosures, approval processes, and documentation.
Change management
We ensure buy-in and clearly defined responsibilities—so that new labeling and approval processes are actually put into practice and are not perceived as an additional bureaucratic burden.
Your expert for AI consulting
Why Choose Ventum Consulting for Quality Management 4.0 Consulting?
: Over 1,500 Projects Completed
Large corporations and small and medium-sized businesses rely on our experience because we deliver what we promise—time and time again.
Over 20 Years of Consulting Expertise at
We know the pitfalls and the shortcuts—so you can get where you’re going faster.
100% Dedicated to Your
Business Success
We aren’t satisfied until you are, because it’s the measurable results that count. That’s how we measure our success.
From AI Strategy to "
" Implementation
Everything from a single source—from the initial potential analysis through model selection to implementation
+1,500 projects completed
Over 20 Years of Consulting Expertise
100% Dedicated to Your Business Success
From AI Strategy to "
" Implementation
- Talk directly with subject matter experts—no sales team involved
- Free Assessment of Your Situation and Needs
Arrange a non-binding initial consultation now
- Legally Compliant: Implementation of Transparency Requirements Under the EU AI Act and Article 50 of the AI Regulation
- Holistic: Integrating AI Governance, Compliance, Content Workflows, and Risk Management
- Future-Proof: Preparing for Regulatory Requirements, Audits, and Competitive Risks
- Proven: Over 20 Years of Expertise in Digital Transformation
- Enablement: Training, Guidelines, and Responsibilities for Marketing, Communications, Legal, and IT
TISAX and ISO certification apply only to the Munich location
Your message
FAQ – Frequently Asked Questions About the AI Disclosure Requirement
The transparency requirements set forth in Article 50 of the EU AI Regulation will take effect on August 2, 2026. For certain technical requirements (e.g., machine-readable watermarks for AI providers), a transition period applies until December 2, 2026.
No. The regulation does not impose a blanket labeling requirement, but rather distinguishes based on the person, purpose, and type of content. For images, videos, and audio, the requirement applies primarily to deepfakes—for text, it applies only to content of public interest that is not subject to human editorial oversight.
A deepfake is any image, audio, or video content generated or manipulated using AI that resembles real people, objects, places, or events and could be mistaken for the real thing. It does not require a famous or identifiable person—even fictional but realistic-looking characters meet the definition.
Generally speaking, no. Traditional product descriptions, marketing copy, and landing pages typically do not fall under the narrow definition of “public interest.” Additionally, the editorial exception applies if your texts are reviewed by an employee before publication.
Generally speaking, no. Color corrections, retouching, the removal of image artifacts, or cropping a photo do not significantly alter the content’s message—in such cases, there is no requirement to label the image. The situation is different, however, if AI suggests a new scene that never actually existed.
Directly on the content or immediately next to it—clearly visible, understandable, and accessible. For images, for example, as a watermark or in a visible caption; for videos, as an on-screen text overlay at the beginning and at regular intervals. A general notice in the legal notice or footer is expressly insufficient.
Purely personal, non-commercial use is expressly exempt from the labeling requirement. So the AI-generated “astronaut uncle” in a family chat remains permitted without any special notation. However, as soon as the content appears in a commercial or public context, the requirement applies.
In principle, the company, as the operator, is responsible. Internally, roles and approval responsibilities should be clearly documented—including for audit trail purposes. In practice, operational responsibilities often fall to marketing, communications, or legal managers.