Chatbot transparency under Article 50
Article 50(1) has applied since 2 August 2026 and places the disclosure duty on the provider of an AI system that interacts directly with people. Run a chatbot, virtual assistant or AI agent under your own name or brand and you can become that provider yourself. Embed AI determines your role per bot, tests whether the exemption holds and delivers the disclosure texts plus the evidence. Lead time: 1 week.
Chatbot transparency
1 week
From an unclear bot to a demonstrable disclosure
Inventory of every bot, assistant and AI agent in customer contact
Per bot: does the disclosure duty apply or does the exemption hold
Concrete disclosure texts for opening message, widget and voice channel
Evidence file with owner, screenshot and date per channel
Do I have to disclose that my customer service bot is AI?
Yes, unless it is clear to a reasonably well informed, observant and circumspect person that they are interacting with AI. Paragraph 1 places that duty on the provider of the system: if you buy a bot in and run it under your own name or brand, or modify it substantially, you can take on that role yourself. Article 50(1) has applied since 2 August 2026 and carries no transition period: your bot has to comply today. The only transition inside Article 50 sits in paragraph 2, applies solely to systems placed on the market before 2 August 2026 and runs until 2 December 2026. For the disclosure duty on direct AI interaction that room does not exist. Organizations that still have to arrange this are late, not early.
When does the disclosure duty fall away because it is obvious?
The exemption is narrow. It applies only where the circumstances themselves make clear that the counterpart is an AI system. A name or a label does not do that work. We test every bot along four questions.
Is it clear to a reasonably well informed and observant person
The benchmark is not your internal knowledge but the perception of an average user in that channel. Doubtful cases fall on the side of the disclosure duty.
A label such as digital assistant is not enough
Terms like digital assistant, virtual colleague or service bot do not meet the threshold. At most they suggest automation, not that the user is talking to an AI system.
A human name and avatar work against you
A bot presenting itself as Lisa from customer service makes the AI nature less clear rather than more. Disclosure then becomes unavoidable and must appear at first contact.
Assess every channel separately
Web widget, WhatsApp, email, telephony and in-app chat differ in what a user sees. The notice visible on your site often does not exist in a voice channel.
Note the difference in timeline: paragraph 1 has applied since 2 August 2026 with no transition period. The transition until 2 December 2026 exists only for the machine-readable marking of paragraph 2 and only for systems placed on the market before 2 August 2026. Regulation (EU) 2026/1744 has been in force since 27 July 2026 and moved high-risk Annex III to 2 December 2027 and Annex I to 2 August 2028, but left Article 50 untouched.
What applies to AI agents acting on your behalf?
Agents that act autonomously, call, email or negotiate require more than a chatbot. The Commission guidelines are explicit here.
Artificial nature and principal
Point 31 of the guidelines requires an AI agent to disclose both its artificial nature and on whose behalf it acts. Stating that it is AI is therefore not enough for an agent.
Your vendor does not solve this for you
Point 117 states that a deployer cannot rely on the machine-readable marking applied by the provider. The visible disclosure in your own channel remains your own duty.
Arrange it contractually in the chain
Point 12 expects proportionate measures in distribution chains, including contractual arrangements. We deliver the clauses to lock this down with your bot vendor or implementation partner.
Outsourcing does not move the duty
Point 14 confirms that a legal person remains the deployer, also where freelancers or an external contact centre carry out the deployment.
Which bots fall within this check?
This check is for any organization with AI in customer contact, whether or not you use high-risk AI elsewhere.
You run a customer service bot
A web widget, WhatsApp bot or in-app chat that answers questions, creates tickets or hands over to an agent.
You deploy a virtual assistant
An assistant in your portal, app or intranet helping customers or citizens, with or without a human name and avatar.
You use AI agents
Agents that autonomously call, email, schedule appointments or prepare transactions on behalf of your organization.
Your bot runs on vendor software
You buy the bot in and assume the vendor handles transparency. Paragraph 1 does sit with the provider, but under your own name or brand you can take on that role yourself and the disclosure has to be visible in your channel.
You have a hybrid human and bot flow
A conversation that starts as a bot and moves to a colleague, or the reverse, where it is unclear who is speaking when.
No high-risk AI, but customer contact
Article 50 is separate from the high-risk regime. Even without an Annex III system your bot can fall under the disclosure duty.
What does the chatbot transparency check deliver?
Inventory of every bot, assistant and AI agent per channel and per brand
Test per bot: does paragraph 1 apply or does the exemption demonstrably hold
Role determination per bot: do you act as provider or as deployer
Ready-to-use disclosure texts in Dutch and English
Placement advice per channel: opening message, widget header, voice intro and handover to a colleague
For AI agents: wording that discloses both the artificial nature and the principal
Contract clauses for your bot vendor and implementation partner
Evidence file with screenshots, owner and date per channel
Action list with owner, priority and lead time
How does the one-week approach work?
Scope and intake
We define which bots, assistants, agents and channels are in scope and who owns each channel.
Channel inventory
We walk through every channel as a user and record what a customer actually sees and hears at first contact.
Test against Article 50(1)
Per bot we determine whether the disclosure duty applies or the exemption holds, and whether you are provider or deployer.
Texts and placement
We deliver the disclosure texts per channel, including the agent wording covering artificial nature and principal.
Evidence and action list
You receive the evidence file per channel plus an action list with owner, priority and lead time.
What does this cost?
Embed AI works with fixed prices for deployers within the confirmed scope, excluding VAT. Providers placing AI systems on the market receive a tailored quote.
Guided AI governance scan
EUR 2,950
Expert assessment of your bots and AI agents, register and first classification. Deductible once if the follow-up starts within 60 days.
AI Act Readiness Sprint
EUR 9,900
Register, classification, gap analysis and a 30, 60 and 90 day roadmap, including the transparency texts and the evidence per channel.
Compliance Bundle
EUR 21,900
The Readiness Sprint plus implementation in policy, contracts, ownership and role-based training for the teams running the bots.
The online transparency scan is free and gives a first picture per bot within minutes. If you then want certainty backed by an expert assessment, the guided scan is the logical step.
Who is this for?
Customer service and customer operations
Teams running the bot day to day who have to implement the disclosures in the conversation flows.
Digital, product and engineering
Teams building the disclosure into widget, app, voice channel and handover to a colleague.
Legal, privacy and compliance
Teams translating the disclosure duty into concrete texts, contract arrangements and demonstrable evidence.
Executives and accountable managers
Who want to know which exposure is open and who closes which action by when.
What do you know afterwards?
Which bots and agents fall under the paragraph 1 disclosure duty
Where the exemption demonstrably holds and where it does not
Which text is shown per channel and who maintains it
How your AI agents disclose the principal they act for
Which evidence you can show a supervisory authority
Logical next steps
Article 50 transparency check
The broader check across all four paragraphs: chatbots, synthetic content, emotion recognition and deepfakes.
View the serviceAI Act readiness and gap analysis
For the full picture of AI systems, roles, duties and evidence in one roadmap.
View the serviceAI inventory setup
For a compact inventory of AI systems with owner, purpose and evidence status.
View the serviceBackground and depth
Legal depth on Praxikon and role-based training at LearnWize for the teams running your bots day to day:
Frequently asked questions
Does every chatbot have to disclose that it is AI?
In principle yes. Article 50(1) requires providers of systems intended to interact directly with natural persons to inform those persons that they are interacting with AI. The only exemption is where this is clear from the circumstances to a reasonably well informed, observant and circumspect person. That exemption is narrow and you must be able to substantiate it per bot and per channel. If you run a purchased bot under your own name or brand, or modify it substantially, you can become the provider yourself and paragraph 1 sits with you.
We call our bot a digital assistant, is that enough?
No. A label such as digital assistant, virtual colleague or service bot does not meet the obviousness threshold. Such terms suggest automation, not that the user is communicating with an AI system. An explicit disclosure at first contact is then required. A human name or avatar makes the disclosure more urgent, not less.
Our chatbot comes from a vendor, do they not handle this?
Not fully. Point 117 of Commission guidelines C(2026) 5054 final of 20 July 2026 states that a deployer cannot rely on the machine-readable marking applied by the provider. The visible disclosure in your own channel remains your task. Point 12 additionally expects proportionate measures in the chain, including contractual arrangements with your vendor. Point 14 confirms that your legal person remains the deployer, also where freelancers or an external contact centre carry out the work.
Is there a transition period for our existing chatbot?
Not for the paragraph 1 disclosure duty. Article 50 has applied since 2 August 2026 and was not postponed by Regulation (EU) 2026/1744, which has been in force since 27 July 2026 and moved only the high-risk regimes. The only transition inside Article 50 sits in paragraph 2, applies solely to systems placed on the market before 2 August 2026 and runs until 2 December 2026.
What is the exposure if the disclosure is missing?
For an infringement of Article 50 a supervisory authority can impose a fine of up to 15 million euro or 3 percent of worldwide annual turnover, whichever is higher. That is a ceiling and not an automatism: the authority weighs nature, duration and gravity. The code of practice on transparency of AI-generated content of 10 June 2026 is voluntary and covers paragraphs 2, 3 and 5, so not the paragraph 1 disclosure duty. Point 148 requires those who do not sign to demonstrate compliance by other adequate means, and point 149 names adherence as a possible mitigating circumstance in a fine.
What does the chatbot transparency check cost?
The online transparency scan is free. For an expert assessment with register and first classification the guided AI governance scan costs EUR 2,950, deductible once if the follow-up starts within 60 days. The AI Act Readiness Sprint costs EUR 9,900 and the Compliance Bundle EUR 21,900, excluding VAT.
Know within a week whether your bots comply.
Start with the free transparency scan. You immediately see per bot whether the paragraph 1 disclosure duty applies. If you want an expert assessment with evidence per channel, plan the gap intake.