Your AI Chatbot Probably Needs a Disclosure Label. The EU Just Explained Exactly What That Means.

The Transparency Rules That Apply From August 2 — Whether You Are Ready or Not

The European Commission published its draft Article 50 transparency guidelines on May 8, 2026, opening a consultation that closes on June 3, 2026.^[1] That is 17 days from today. If your business runs AI that interacts with European customers, employees, or the public, this consultation should be on your radar — and here is why.

Article 50 of the EU AI Act is one of the few obligations in the regulation that applies before the high-risk regime kicks in. From August 2, 2026, anyone deploying AI that speaks to a human, generates content, or processes biometric data in certain ways must disclose it — clearly, at the point of first interaction. The Commission's draft guidelines are the first formal interpretation of exactly what that means in practice.^[3]

The guidelines are non-binding. But they are also the only authoritative interpretation of Article 50 that exists. National regulators will use them. Courts will cite them. If you are not part of the consultation, you will be on the receiving end of whatever the final text says — and you will have had no input into shaping it.

What Article 50 Actually Requires — And Who It Applies To

Article 50 breaks into four distinct obligations, depending on what your AI system does:^[2]

• Article 50(1) — AI interaction disclosure (providers): If you place an AI system on the EU market that is designed to interact with natural persons — a chatbot, a virtual assistant, an AI companion — you must inform users that they are talking to an AI. This applies to providers, not deployers. The obligation is embedded in the product.
• Article 50(2) — Machine-readable marking (providers): If your AI generates or manipulates content — images, audio, video, text — you must embed machine-readable marks that enable detection of that content as AI-generated. This is the watermarking obligation, now due December 2, 2026 per the Omnibus deal.^[1]
• Article 50(3) — Emotion recognition and biometric disclosure (deployers): If you deploy a system that infers emotional state or categorises individuals based on biometric data — voice analysis, facial expression scoring, stress detection — you must inform the exposed person before the system operates. This is a deployer obligation. The August 2, 2026 deadline has not moved.
• Article 50(4) — Deepfake and AI-generated text labelling (deployers): If you publish AI-generated content on matters of public interest — news articles, political content, public health information — or if you expose someone to a deepfake, you must label it. The Omnibus extended the labelling deadline to December 2, 2026. The machine-readable marking piece got the same extension. Human-readable labelling may still follow the August 2 timeline depending on how the final Omnibus text is interpreted.^[3]

The guidelines cover all four. They are the first document that tells providers and deployers — in concrete, operational terms — how each obligation is supposed to work.

The "Obviousness" Exception: When You Do Not Have to Disclose

Here is the part most businesses will focus on first. Article 50(1) does not apply where the AI nature of the interaction is "obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect." The guidelines adopt the EU consumer law "average consumer" standard — not the most naive user, not the most sophisticated one.^[2]

The Commission then gets specific about where the exception applies — and where it does not.^[3]

Probably obvious — no disclosure required:

• AI-powered code assistance chatbots available only to professional developers (the audience is technically sophisticated and the tool's purpose makes AI obvious)
• AI-enabled NPCs in video games — the guidelines conclude the gaming context makes AI interaction obvious to the reasonably observant player

Probably not obvious — disclosure required:

• AI-enabled robotic companion pets designed to mimic natural human-pet interaction — the animal-like interface obscures the AI nature of the interaction
• AI avatars in immersive VR/AR environments where the experience is designed to feel human
• Chatbots embedded in online helpdesks where the visual design mimics a human support agent
• Any AI system where the target audience includes children, elderly persons, or persons with disabilities — the guidelines require a higher bar for obviousness when vulnerable groups are likely users

This matters practically: most businesses running a chatbot on their website cannot rely on the obviousness exception. The guidelines specifically call out the helpdesk scenario. If your chatbot looks like a person and your customer might be elderly, vulnerable, or not technically literate, you need a disclosure.

The Agentic AI Complication Nobody Planned For

The guidelines address a scenario that is moving up the regulatory agenda fast: AI agents. The question is whether a system that acts autonomously on a user's behalf — booking flights, sending emails, making purchasing decisions — falls within Article 50(1) when it interacts with third parties on behalf of the user.^[2]

The Commission's answer: yes, where the agent is designed to interact with natural persons. And the critical part — where the provider cannot reliably determine in advance whether the agent will interact with a natural person, the agent should be instructed to disclose its AI nature in every situation where such interaction is likely.^[3]

If you are building or deploying AI agents in the EU, this is a direct compliance requirement that applies from August 2, 2026. Your agent needs to be capable of disclosing itself as AI when it contacts a human. This is a system design requirement, not a policy add-on.

The Disseminator Carve-Out — And Why Online Platforms Are Paying Attention

The guidelines include a significant carve-out on the deployer side. Actors whose role is limited to transmitting or disseminating AI-generated content created by a third party — online platforms, social media networks, content aggregators — are not deployers under the AI Act, because they do not exercise "authority" over the AI system that produced the content.^[2]

This matters for the AI-generated text labelling obligation under Article 50(4). If an online platform is not a deployer, it is not legally required to apply the labelling that the upstream provider should have applied. The platform can rely on the upstream marking.

That said, the guidelines encourage platforms to "preserve the machine-readable marks and labels applied upstream" and to take "appropriate measures" to ensure individuals exposed to AI-generated content are informed of its artificial origin. The obligation is softer for disseminators — but the Commission has made clear it expects best efforts anyway.^[3]

The Consultation Closes June 3 — And You Probably Have Something to Say

If your business operates AI systems that interact with EU residents, the consultation is worth engaging with. The guidelines are the first and only formal interpretation of Article 50. The final text will shape how national regulators — who are responsible for enforcement under the AI Act — understand your obligations.^[1]

Specific areas where feedback could be most valuable:

• The obviousness test: If your AI system operates in a context where you believe disclosure is unnecessary because the AI nature is self-evident, the guidelines provide your defence — but only if your specific use case fits within the examples given or clearly meets the "average consumer" standard. If it does not, submitting feedback on where the guidelines are too narrow is worth doing before June 3.
• Agentic AI obligations: The requirement that agents self-disclose in every likely human interaction is a significant design constraint. If this is technically challenging for your system architecture, the consultation is the formal channel to raise it.
• Disseminator obligations: If you operate a platform that publishes third-party AI-generated content, the distinction between "preserving upstream marks" and an independent labelling obligation may affect your compliance process. If the current drafting is ambiguous for your use case, flag it.

The consultation is targeted and structured. Providers, deployers, businesses of all sizes, public authorities, academics, and citizens can respond. The Commission will consider all submissions before adopting the final guidelines, expected alongside the August 2 application date.^[1]

What to Do in the Next Six Weeks

The deadline is August 2, 2026. You have six weeks after the consultation closes before the transparency obligations apply. Here is a practical sequence:

• Audit your AI systems now against Article 50 categories. Which of your systems falls within 50(1), 50(2), 50(3), or 50(4)? Identify the applicable obligation and the relevant disclosure requirement for each.
• Assess your obviousness defence for every chatbot or interactive AI. If your chatbot looks like a human agent and your audience includes anyone other than technically sophisticated professionals, assume you need a disclosure. The guidelines give you the arguments for why you might not — but the burden of proving obviousness is on you.
• Design your agent disclosure protocol if you deploy AI agents. Your agent needs to be able to identify when it is likely to interact with a natural person and disclose its AI nature. This is a system design requirement that needs to be built in, not added on.
• Engage with the consultation if the guidelines are ambiguous for your use case. June 3 is the deadline. Your submission does not need to be comprehensive — a focused response on the specific provision that affects your business is valuable.

The Bottom Line

Article 50 is not the high-risk regime. There is no conformity assessment, no notified body, no technical documentation audit required. But it is a legal obligation that applies to every business placing interactive AI, generative content tools, or biometric processing systems on the EU market — and the August 2 deadline is real.

The Commission's draft guidelines are the first detailed interpretation of what you actually have to do. Read them with your product team, not just your legal counsel. The obviousness test is ultimately a factual question about how your system appears to the people who use it. That is a product and design question as much as a legal one.

And if the guidelines miss something that matters for your use case, submit to the consultation before June 3. Seventeen days is not much time. But it is still time.

This article is for informational purposes only and does not constitute legal advice.