EU AI Act transparency requirements are now in force: What this means for communications and content production

2 August 2026 is a historic date in the European AI landscape. The EU AI Office and national supervisory authorities have officially begun enforcing the AI Act, and the transparency requirements under Article 50 have entered into force at the same time.

Breaches can lead to substantial penalties: fines may reach up to €15 million or 3% of a company’s worldwide annual turnover. The regulation applies to all organisations that use or provide AI systems within the EU.

01EU AI Act timeline: what applies now?

The AI Act applies in stages. The Digital Omnibus package adopted in 2026 postponed high-risk obligations, but the Article 50 transparency requirements stayed on schedule. The key dates for communications:

• 1 August 2024: the AI Act entered into force.
• 2 February 2025: prohibited AI practices and Article 4 on AI literacy began to apply.
• 2 August 2025: obligations for providers of general-purpose AI (GPAI) models began to apply.
• 2 August 2026: Article 50 transparency requirements began to apply: disclosing AI interaction, disclosing deepfakes and labelling text published on matters of public interest.
• 2 December 2026: the grace period ends for machine-readable marking in generative systems placed on the market before 2 August 2026.
• 2 December 2027: high-risk (Annex III) obligations begin to apply after the Omnibus postponement.

Always check the current position in the official legal text and the Commission's guidance. The Article 50 guidelines are still being finalised.

02Article 50: Four key transparency requirements

The new rules focus on people’s right to know when they are dealing with AI or AI-generated content. The requirements fall into four areas:

• Interaction with AI: Chatbots, AI assistants and similar systems must be designed so that users immediately know they are interacting with AI, not a human.
• Synthetic content: AI-generated images, video and audio (including deepfake content) must carry a machine-readable mark so they can be identified as AI-made. For generative AI systems placed on the market before 2 August 2026, a transition period runs until 2 December 2026.
• Emotion recognition and biometric categorisation: Where these techniques are used, the people affected must be informed.
• AI-generated text of public interest: If an organisation publishes AI-generated text on topics of societal importance or public interest, this must be disclosed openly — unless the content has undergone an appropriate editorial process. This often matters especially for public-sector communications.

03Editorial responsibility lifts the labelling obligation

This last point is especially critical for communications teams. Under the regulation, the labelling obligation does not apply if the published text has undergone substantive human review and a natural or legal person holds editorial responsibility for it.

In practice, using AI for drafting or brainstorming does not require a label next to the published text — as long as a human stays in control (human-in-the-loop) and the approval chain is verifiable. If text is published directly without human review and editing, you enter a legal grey area.

AI does not replace the communicator. Instead, the communicator’s role shifts from mechanically writing routine copy towards strategic editing, quality assurance and publication coordination.

04Proof of Value: How Lyyli supports compliance

In many communications teams, AI use is fragmented. Ideas and drafts bounce around separate ChatGPT or Claude windows, and final approvals are chased by email or chat. That way of working leaves no documented trail of who approved the text and what human review it received.

Lyyli is not just a text generator. It is a full communications operations system. It addresses the challenges of the AI Act by bringing the entire process into one managed view:

• Documented approval process (Audit Trail): For every draft, you can see exactly who created, edited and approved it. This provides strong evidence that a human holds editorial responsibility for the published content.
• Secure, professional working environment: Lyyli is built for organisational use with security in mind. Customer data is protected with encryption in transit and at rest, and workspaces for different organisations are kept strictly separate. We process data in line with the GDPR and agreed terms.
• Transparent AI routes: On selected AI routes, we ensure customer-submitted content is not used to train language models and is not permanently stored in AI partners’ systems. We explain this and our data-handling principles openly on our Trust page.
• Brand control and consistent guidelines: Lyyli ensures AI always follows the organisation’s own brand guidelines and tone of voice. That helps prevent misleading or inconsistent communications.

Custom basic workflows or standalone chatbot windows cannot deliver the transparency and security a modern organisation needs in the new era that began on 2 August 2026. With a managed workflow, your communications stay secure, efficient and compliant.

05Checklist: what communications teams should do now

• Map every place AI is used: drafting, images, video, audio, chatbots and translation.
• Agree who carries editorial responsibility for each content type and how approval is documented.
• Label AI-generated images, video and audio that look like real people, places or events (deepfakes).
• Tell users clearly in your website chatbot or customer service that they are interacting with AI.
• Write AI-use principles into your communications guidelines and train the team (Article 4, AI literacy).
• Check how the AI services you use mark their output in a machine-readable way.

06How this shows up in the communications team's daily work

In practice, Article 50 does not demand a new layer of bureaucracy. It demands a verifiable workflow. Needs, drafts, comments and approvals stay in one view; AI helps draft from the organisation's brand guidelines, but people decide what is published. Documented approval leaves an audit trail, and approved content is taken to the destination platform. When that is part of everyday work, the labelling obligation usually does not arise — because editorial responsibility is already demonstrable.

07Frequently asked questions about the EU AI Act in communications

Is the EU AI Act already in force?

Yes. The regulation entered into force on 1 August 2024, and its obligations apply in stages. The Article 50 transparency requirements, which matter most for communications, have applied since 2 August 2026.

Does every AI-assisted text need a label?

No. The text-labelling obligation covers AI-generated text published to inform the public on matters of public interest. It does not apply when the text has undergone human review and someone holds editorial responsibility for it.

What about AI-generated images?

If an image, video or audio resembles real people, places or events and could appear authentic, it must be disclosed (deepfake). Obviously artistic or creative use carries a lighter duty. Human approval does not remove the disclosure duty for images in the same way it does for text.

What happens if a company does not comply?

Breaching Article 50 can lead to a fine of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. A lower cap applies to SMEs.

What did the Digital Omnibus change?

It mainly postponed high-risk obligations and gave existing generative systems until 2 December 2026 for machine-readable marking. The other Article 50 transparency requirements already apply.

08Sources

Legal text: EUR-Lex: AI Act (EU) 2024/1689