From chatbots to deepfakes and AI-generated content, the EU’s new transparency requirements are now operational. For businesses, the priority is to understand which rules apply, who is responsible and where disclosure is actually required.
Artificial intelligence has become embedded in customer service, marketing, communications, content production and internal workflows.
On 2 August 2026, another important part of the European Union’s AI regulatory framework became applicable: the transparency obligations in Article 50 of the EU Artificial Intelligence Act.
These requirements are designed to help people recognise when they are interacting with an AI system or encountering certain types of AI-generated or manipulated content. They apply differently depending on whether an organisation is acting as a provider or a deployer of an AI system and on how the technology is being used.
For businesses, this distinction matters.
The AI Act does not create a single rule stating that every use of generative AI must be disclosed. Instead, Article 50 establishes specific transparency obligations for particular systems and use cases.
That means companies need more than a generic “AI-generated” label.
They need to understand where AI is being used, what role they play under the Regulation and which transparency obligation applies.
Here are five areas businesses should address now.
1. Identify where people are interacting directly with AI
One of the most visible requirements concerns AI systems that interact directly with people.
Under Article 50(1), providers must ensure that AI systems intended to interact directly with natural persons are designed and developed so that those individuals are informed that they are interacting with an AI system.
There is an exception where this would already be obvious to a reasonably well-informed, observant and circumspect person in the circumstances. The European Commission has indicated that this exception should be interpreted restrictively.
Examples of potentially relevant systems include:
- customer-service chatbots;
- conversational AI assistants;
- AI agents communicating directly with customers;
- virtual avatars;
- other interactive AI interfaces that communicate directly with individuals.
The Commission’s guidance explains that the rule applies where there is a genuine direct interaction between the AI system and a natural person. Systems operating only in the background or through machine-to-machine communication do not fall within this particular disclosure obligation.
What businesses should do
Create an inventory of customer-facing and employee-facing interactive AI systems.
For each system, establish:
- who provides the AI system;
- whether users interact directly with it;
- whether the AI nature of the interaction is already obvious;
- how and when the disclosure is presented;
- whether the information is accessible and clearly distinguishable.
The Commission states that people should be informed from the beginning of the first interaction where the requirement applies.
A company should therefore not wait until the end of an AI-assisted conversation to explain that AI was involved.
2. Understand the difference between visible disclosure and machine-readable marking
Article 50 contains two concepts that businesses should not confuse.
One concerns what people can see or understand.
The other concerns what machines can detect.
Providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content must ensure that relevant outputs are marked in a machine-readable format so that they can be detected as artificially generated or manipulated.
The Regulation requires these technical solutions to be effective, interoperable, robust and reliable as far as technically feasible, taking account of factors including the type of content, implementation costs and the state of the art.
However, this does not mean that every AI-assisted edit automatically falls within the marking requirement.
Article 50 provides an exception where an AI system performs an assistive function for standard editing or does not substantially alter the input data or its meaning. The Commission’s guidance also provides further examples and clarification concerning outputs that fall outside the obligation.
This is particularly relevant for companies using AI-enabled creative, marketing and productivity tools.
What businesses should do
When procuring or deploying generative AI systems, ask providers:
- Does the system generate synthetic text, audio, images or video?
- Does it implement the machine-readable marking required by Article 50?
- What technical method is used?
- Is the marking preserved when content is exported?
- Are there workflows that may remove relevant metadata or other markings?
- Which outputs are treated as outside the Article 50(2) requirement?
Procurement and technology teams should make these questions part of AI vendor due diligence rather than leaving them exclusively to marketing or legal departments.
3. Establish a clear policy for deepfakes
Deepfakes receive specific treatment under Article 50.
The AI Act defines a deepfake 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.
Where an organisation acting as a deployer uses an AI system to generate or manipulate image, audio or video content that constitutes a deepfake, Article 50(4) requires disclosure that the content has been artificially generated or manipulated.
The disclosure must be clear and distinguishable and must be provided no later than the person’s first exposure to the content. The Commission has specifically explained that a deployer cannot rely solely on an invisible machine-readable mark to satisfy this human-facing disclosure requirement.
There is a tailored rule for content forming part of evidently artistic, creative, satirical, fictional or analogous works or programmes. In such cases, disclosure is still required, but it may be provided in an appropriate manner that does not interfere with the display or enjoyment of the work.
What businesses should do
Marketing, communications and creative teams should establish a review process for AI-generated or AI-manipulated:
- photographs;
- advertisements;
- promotional videos;
- synthetic voices;
- virtual presenters;
- realistic representations of people, products, locations or events.
The key question is not simply:
“Was AI used?”
It is:
Does the content meet the AI Act definition of a deepfake, and if so, how must it be disclosed?
That distinction prevents both under-disclosure and unnecessary blanket labelling.
4. Treat AI-generated public-interest text differently from ordinary business content
Another frequently misunderstood provision concerns AI-generated text.
Article 50 does not say that every email, product description, social-media post or marketing text created with AI must automatically carry an AI label.
The specific disclosure obligation applies to AI-generated or manipulated text that is published for the purpose of informing the public on matters of public interest.
The Commission identifies examples of public-interest subject matter including areas such as:
- politics and democratic processes;
- public administration;
- justice and law enforcement;
- fundamental rights;
- public security;
- public health;
- environmental protection;
- consumer safety;
- economic, financial, scientific and cultural developments relevant to public debate.
There is also an important exception.
Disclosure is not required under this provision where the AI-generated text has undergone a process of human review or editorial control and a natural or legal person holds editorial responsibility for its publication.
The Commission further clarifies that meaningful human review involves substantive examination by a person with relevant knowledge and professional judgement. Merely checking spelling or grammar is not enough to constitute the type of human review described in its guidance.
This distinction is highly relevant for corporate publishing.
What businesses should do
Establish an editorial workflow for AI-assisted public communications.
For higher-risk or public-interest material:
- Identify whether generative AI was used.
- Verify factual claims against authoritative sources.
- Conduct substantive human review.
- Record who approved the content.
- Establish who holds editorial responsibility.
- Determine whether an Article 50 disclosure is required.
This approach is stronger than simply adding “Created with AI” to everything a company publishes.
It creates accountability.
5. Build AI transparency into governance and procurement
Article 50 is not merely a communications issue.
It affects AI governance, vendor management, product development, procurement, marketing and compliance.
A company may use dozens of AI-enabled systems across different departments without maintaining a central record of where AI is interacting with people or producing content.
That creates a basic governance problem:
An organisation cannot apply the correct transparency rule if it does not know where AI is being used.
Businesses should therefore incorporate transparency into their wider AI governance framework.
A practical inventory could record:
| Question | Example |
|---|---|
| What AI system is being used? | Customer-service assistant |
| Who provides it? | External software provider |
| Who deploys it? | Company customer-service team |
| Who interacts with it? | Customers |
| What does it generate? | Conversational text |
| Does Article 50 apply? | Assessment required |
| What disclosure is used? | AI interaction notice |
| Who owns compliance? | Product / Legal / Compliance |
| Is evidence retained? | Yes / No |
This type of register can also support procurement.
Before approving a new AI system, procurement teams should ask vendors about:
- Article 50 functionality;
- machine-readable marking;
- transparency settings;
- documentation;
- technical limitations;
- downstream responsibilities;
- changes to compliance functionality;
- support for applicable accessibility requirements.
The result should be a chain of accountability from:
AI system → provider → business use case → user → output → transparency requirement → responsible owner
A limited transition period applies—but only in a specific case
Businesses should also understand the transition rule correctly.
Article 50 generally applies from 2 August 2026.
However, the European Commission states that a limited grace period applies to AI systems placed on the market before 2 August 2026 specifically in relation to the Article 50(2) machine-readable marking and detection obligation for AI-generated content.
Providers of those systems must comply with that particular obligation from 2 December 2026.
The Commission makes clear that this is a limited transition period rather than a general postponement of Article 50.
Companies should therefore not treat December 2026 as the general start date for AI transparency compliance.
It is not.
What businesses should do in the next 30 days
For organisations that have not yet operationalised Article 50, five immediate actions are sensible.
1. Map AI use cases
Identify interactive AI systems, generative AI tools, synthetic-media workflows and relevant biometric or emotion-recognition systems.
2. Determine legal roles
Establish whether the organisation is acting as a provider, deployer or another actor for each relevant AI system.
3. Audit disclosures
Check whether required AI interaction notices and content disclosures are clear, distinguishable, timely and accessible.
4. Review vendors
Confirm how AI suppliers implement machine-readable marking and other Article 50 functionality relevant to their systems.
5. Establish editorial responsibility
Create a documented human-review process for AI-assisted content on matters of public interest.
Transparency should become part of AI design—not an afterthought
The practical significance of Article 50 extends beyond labels.
It introduces a straightforward principle into enterprise AI adoption:
People should be able to understand when certain consequential interactions or content involve artificial intelligence.
For organisations, implementing that principle requires coordination.
Technology teams need to understand how systems work.
Procurement teams need to know what vendors provide.
Marketing teams need rules for synthetic media.
Editors need standards for AI-assisted publishing.
Compliance teams need to understand where disclosure is legally required.
And leadership needs visibility across the organisation.
The companies that manage this through isolated warnings and ad hoc labels are likely to create inconsistency.
Those that build transparency into AI governance from the beginning will be better positioned to adopt new AI tools while maintaining accountability and trust.
The X3AI perspective
The latest phase of the EU AI Act illustrates an important shift in enterprise AI.
The challenge is no longer simply deciding whether to use artificial intelligence.
The challenge is using it responsibly, transparently and operationally at scale.
That requires more than a policy document.
Organisations need practical governance, employee awareness, appropriate technology controls and clear processes connecting AI systems to real business use cases.
AI transparency should therefore be treated not as a barrier to adoption, but as part of mature AI adoption itself.
X3AI — Enabling people and organisations to turn AI into practical value.
Primary sources
European Commission, “Guidelines on transparency obligations for providers and deployers of AI systems,” published 20 July 2026.
European Commission, “Transparency obligations under Article 50 of the AI Act” — official questions and answers.
European Commission, “Quick Facts: Transparency rules for AI systems.”
Regulation (EU) 2024/1689 — Artificial Intelligence Act, Article 50, official consolidated text on EUR-Lex.
European Commission, “Code of Practice on Transparency of AI-generated Content.”
This article is provided for general informational purposes and does not constitute legal advice. Organisations should assess the Artificial Intelligence Act in light of their specific systems, legal roles and use cases.
Schreiben Sie einen Kommentar