EU AI Act Transparency Rules Are Now in Force: What Businesses Need to Review
The EU AI Act's transparency rules now apply. For businesses using customer-facing AI, the practical question is simple: are people clearly told when they are interacting with AI or seeing synthetic content?
Picture a website visitor starting a conversation with a helpful assistant. The answers are quick, the tone is natural, and the visitor assumes a person is on the other side. That small moment now has a regulatory dimension for businesses serving people in the European Union.
On August 2, 2026, the transparency rules in Article 50 of the EU AI Act began to apply, and enforcement began for the provisions already in scope. The rules are intended to help people recognise when they are interacting with AI or viewing AI-generated or manipulated content. [1] [2]
This does not mean every company using an internal AI tool needs to build a large compliance department overnight. It does mean that businesses should take a fresh look at customer-facing chatbots, AI agents, avatars, synthetic media, and automated public content. The practical question is straightforward: can a reasonable person tell when AI is involved?
The Update in Plain English
Article 50 is about transparency. It separates the responsibilities of companies that provide AI systems from the responsibilities of organisations that use them professionally.
For people who interact directly with an AI system, such as a chatbot, AI agent, or avatar, the system must be designed so people are informed they are interacting with AI unless that fact is obvious. The European Commission says the notification should appear from the start of the first interaction in a clear and distinguishable way. [3]
The update also addresses specified synthetic content. Qualifying deepfakes must be clearly labelled. Certain AI-generated or manipulated text published to inform the public on matters of public interest must also be labelled when it has not received human review and editorial responsibility. Providers of generative AI systems have separate duties around machine-readable marking and detectability of generated or manipulated output. [4]
Why a U.S. Business Should Pay Attention
Geography is not always a clean boundary in digital services. The Commission's Article 50 FAQ states that providers outside the EU can be subject to the AI Act where the output of their AI system is used in the EU. A U.S.-based company with EU customers, visitors, clients, or a product offered into the EU should not assume that its location settles the question. [3]
The first step is not panic. It is clarity. A business needs to know which AI experiences it operates, who sees them, and whether it is acting as the provider, the professional user of a vendor's system, or both. The exact legal analysis depends on the product, role, use case, and markets served. Where there is a meaningful EU connection or material risk, involve qualified legal counsel.
The Situations Worth Reviewing First
A Website Chatbot, AI Agent, or Digital Avatar
People should be informed when they are directly interacting with AI unless that fact is obvious. A practical first step is a clear opening disclosure, such as: “You are chatting with an AI assistant. It can help with general questions and connect you with our team.” [3]
AI-Created Video, Audio, or Imagery That Could Appear Authentic
Deepfakes that meet the legal definition require clear labelling. Build a review and labelling step before publishing synthetic media, especially when it depicts people, voices, events, or places in a way that could appear authentic. [1]
AI-Written Public Information Without Human Review
Text produced or manipulated by AI for publication on matters of public interest can require a clear label when it has not gone through human review and editorial control. The operational response is simple: make named human review and editorial sign-off part of your publishing process, or apply the appropriate disclosure where required. [4]
Internal Tools Running in the Background
The direct-interaction disclosure rule is less likely to be the central issue for internal systems that do not directly communicate with people. The Commission distinguishes background and machine-to-machine systems from genuine two-way exchanges with people. That does not remove the need for sensible governance. Keep an inventory, document the system owner, and review data, security, and broader risk controls. [3]
Provider and Deployer: The Distinction That Matters
The law separates the company providing an AI system from the organisation using it under its authority. A provider develops an AI system, or has it developed, and places it on the market or puts it into service under its own name or trademark. A deployer is a person or organisation using an AI system in a professional activity. [3]
For a business using a third-party chatbot platform, the vendor may have provider responsibilities. But the business should not stop there. Confirm how the vendor handles transparency design and content marking, then make sure the customer experience is clear, accurate, and consistent with your own policies.
A Practical 30-Day Response Plan
Start with a simple inventory. List every AI experience your customers, prospects, employees, or partners can see. Include chatbots, agents, automated email responses, AI voice tools, generated videos, image tools, and public-facing automated content.
Next, identify an owner for each use case. Someone should be accountable for keeping the disclosure current, reviewing vendor changes, and escalating questions that have legal, privacy, security, or reputational consequences. Then test the experience as a first-time visitor. Is the AI disclosure present from the start? Is it understandable on a phone? Does the user have a clear way to reach a person when needed?
Document what you find. Screenshots of the disclosure, a record of the tool, its owner, its purpose, and a vendor contact are useful evidence that the business approached the issue thoughtfully. Where a vendor provides compliance documentation or information about machine-readable marking, keep it with the system record.
Finally, build transparency into your approval process. Every new customer-facing AI tool should have an owner, a visible disclosure plan, and a short review before it goes live. That is much easier than untangling dozens of AI experiments later.
Keep the Timeline in Perspective
The AI Act applies in stages. The transparency rules are now live. The current official timeline schedules rules for certain Annex III high-risk AI systems from December 2, 2027, while rules for high-risk AI embedded in regulated products are scheduled from August 2, 2028. [2]
That gives most small and mid-size businesses time to build a sensible, proportionate governance foundation rather than react under pressure. The immediate opportunity is straightforward: make customer-facing AI easy to identify, keep a record of what you use, and make sure someone is responsible for the experience.
The Bottom Line
The EU AI Act's Article 50 rules are a reminder that AI governance begins long before a company deploys a high-risk system. It begins with honest communication about when AI is involved. If your business uses AI to interact with customers, generate content, or support an EU-facing service, now is a good time to review those touchpoints.
This article is general information, not legal advice. The applicability of the EU AI Act depends on the specific system, role, use case, and markets involved.
Sources
[1] European Commission, “Safer and more transparent AI,” August 2, 2026
[2] European Commission AI Act Service Desk, “Timeline for the Implementation of the EU AI Act”
[3] European Commission, “Transparency obligations under Article 50 of the AI Act”
[4] European Commission, “Code of Practice on Transparency of AI-generated Content”
Have a question about this article?
Chat with our team directly on WhatsApp.
Pedro Prado
Founder, Prado AI Solutions. 30+ years in operational risk, audit, and controls. Published author on AI. Trilingual: English, Spanish, Portuguese.