EU AI Act Article 50 Sales Rules: When AI Must Say It's AI

What does Article 50 of the EU AI Act require?

Article 50 sets four separate transparency duties, and only some of them touch sales. The one that matters most for outreach is paragraph 1 of the AI Act, Regulation (EU) 2024/1689: providers must design AI systems “intended to interact directly with natural persons” so that the people concerned are informed they are interacting with an AI system. The duty falls away when that is obvious to a reasonably well-informed, observant and circumspect person, given the circumstances and context. Paragraph 5 adds that the information must be clear and distinguishable and given at the latest at the first interaction or exposure.

Article 3 defines the two roles. 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 uses an AI system under its authority, unless the use is a personal, non-professional activity.

When did Article 50 start to apply, and did the omnibus move it?

Article 50 has applied since 2 August 2026, and the 2026 changes kept that date for the disclosure duty. Article 113 of the AI Act sets 2 August 2026 as the general application date and does not list Article 50 among its exceptions. The Commission's Article 50 questions and answers, last updated 24 July 2026, confirm the date and describe one limited grace period.

That grace period comes from the AI Omnibus, Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026. It adds a paragraph to Article 111: providers of generative AI systems placed on the market before 2 August 2026 have until 2 December 2026 to comply with the marking duty in Article 50(2). The omnibus also rewrote Article 50(7), on codes of practice for marking and labelling AI-generated content. Law firms read the changes the same way: Gibson Dunn wrote on 27 May 2026 that the duty to tell users they are interacting with AI was unaffected, and Jones Walker wrote on 16 July 2026 that most Article 50 deadlines stayed at 2 August.

The penalty ceiling for breaching Article 50 is in Article 99(4)(g): up to €15 million or 3% of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs, including start-ups, Article 99(6) applies whichever of the two is lower; the omnibus's new Article 99(6a) does the same for small mid-caps. Member states lay down the penalty rules under Article 99(1), and the Commission's Q&A says national market surveillance authorities will mainly enforce Article 50.

Does an AI agent that messages prospects count as interacting directly?

The Commission's guidelines say AI agents are covered when they can interact with people while carrying out tasks such as managing correspondence, and their description of direct interaction, typically real-time or near real-time, includes a system's capability to write and send messages to people. The guidelines, approved on 20 July 2026 as C(2026) 5054, list four elements for Article 50(1): an AI system, intended to interact, directly, with natural persons. Their explanation of each element is where outreach tools come in:

• An AI system. Traditional out-of-office emails and rule-based quick replies are not AI systems, so they are outside the duty.
• Intended to interact. The system must be capable of a two-way exchange with a genuine conversational or responsive character. The interaction does not have to be started by a human, may be AI content that draws a human response, can be a single exchange, and the person's input can arrive by e-mail.
• Directly. This typically means real-time or near real-time interaction, and the guidelines include a system's capability to write and send messages to people.
• With natural persons. Professionals count. Machine-to-machine traffic that is not meant to reach a person does not.

Paragraph 31 of the guidelines deals with agents. AI agents are covered if they can interact with people while executing tasks, with making bookings and managing correspondence among the examples, and they must be designed and developed to disclose both that they are AI and on whose behalf they act. One disclosure technique the guidelines list is an email generated by an AI agent that carries an AI label at the top. The in-scope examples include chatbots and conversational agents in customer support and e-commerce, voice assistants, and AI bots on social networks.

The Article 50(1) duty sits with the provider. The guidelines' example is a company that offers a chatbot or AI agent under its own name to consumers, professional users or other organisations that may fit it into their existing processes: that company is the provider. A company that builds its own chatbot in-house and runs it under its own name is a provider too.

What if a person reviews and sends the AI's draft?

The guidelines treat mediated interaction, such as customer service staff using AI assistance tools to help them communicate, as indirect and outside Article 50(1), with two limits. Direct interaction, they say, excludes cases where someone sees AI output without interacting with the system, including AI output passed on by another person. But the mere possibility that a human could step in or review the output should not be used to avoid disclosure. And products that blend AI responses with human content need disclosure for the AI parts, unless those outputs have been “properly reviewed and sent by humans as the main interlocutors”.

Two other paragraphs come up in the same conversations. The text-labelling duty in Article 50(4) covers text published to inform the public on matters of public interest, and the guidelines list private, interpersonal correspondence for professional purposes as an example of text that is not published. The marking duty in Article 50(2) applies to providers of generative systems, including those producing text. It does not cover assistive standard editing such as grammar correction, spellchecking or translation, and its narrow business-to-business exemption covers only strictly technical, internal output not meant to be shared outside the company.

For how teams split drafting and sending between AI and people, see our guide to human-in-the-loop AI outreach.

What is still unsettled?

Several questions that matter for outbound sales have no settled answer yet. These are the ones we found while reading the texts:

• The guidelines are not binding. They say so themselves, and add that only the Court of Justice of the EU can give an authoritative interpretation of the AI Act.
• There is no sales example. We searched the 51-page English text of the guidelines for outreach, prospect, sales and cold and found none. Applying the agent and email passages to cold outreach is a matter of reading, without a worked example.
• How much review is enough. The Article 50(1) section says AI outputs properly reviewed and sent by humans fall outside, but sets no test for proper review. The definition of human review in the guidelines sits in the Article 50(4) section, for public-interest text.
• One-way first messages. The guidelines require a capability for two-way exchange, while also saying a single exchange can count and the AI can start it. They do not address directly a system that only sends a first message while people handle every reply.
• Obviousness with business buyers. A professional audience can make AI more obvious, the guidelines say, but the exception is read restrictively and limited to cases with almost no doubt. Writing patterns and the capability for advanced personalised interaction are among the factors that can affect the assessment.
• Reach outside the EU. Article 2 covers providers placing AI systems on the EU market or putting them into service there, wherever they are based, deployers in the EU, and providers or deployers elsewhere when the AI output is used in the EU. Jones Walker says mere accessibility from the EU is not by itself the trigger and the analysis is role- and fact-specific.

What should a sales team do this week?

Map where AI meets your buyers, ask your vendors how they handle Article 50, and take advice before the next campaign. These steps are our suggestions, not requirements from the regulation:

1. List every AI touchpoint with buyers

   Website chat, voice agents, AI agents that send messages, and writing assistants. For each, note whether the AI sends on its own or a person reviews and sends. If you use Siela's AI SDR, record how it works: each LinkedIn and X message is written by AI for that person from their profile and the signal that surfaced them, checked by a second AI reviewer, and sent on autopilot once you press Save and start.

   One sheet, one hour

2. Ask each vendor about 50(1) and 50(2)

   Ask how their system tells people they are dealing with AI and how its outputs are marked. In the guidelines' examples, these duties sit with the company that offers the system under its own name.

   Vendor email

3. Decide where a disclosure would appear

   Article 50(5) asks for clear and distinguishable information at the first interaction. The guidelines say information buried in a manual, menus or terms of use that people rarely read does not count as clear.

   First contact

4. Write down any human review you rely on

   Record who reviews, what they check and who presses send. Jones Walker recommends documenting the facts behind any conclusion that an exception applies.

   Keep a record

5. Read the Commission's texts, then take advice

   Start with the Commission's Q&A page. Section 3 covers interactive systems; section 7 covers how and when to inform people. Bring your touchpoint list to a lawyer who knows EU law.

   Before the next campaign

Frequently asked questions

Sources

1. EUR-Lex: Regulation (EU) 2024/1689 (AI Act) (checked 20 Oct 2026)
2. EUR-Lex: Regulation (EU) 2026/1744 (Digital Omnibus on AI) (published 24 Jul 2026)
3. European Commission: Guidelines on transparency obligations for providers and deployers of AI systems (published 20 Jul 2026)
4. European Commission: Guidelines text, Annex to C(2026) 5054 final (PDF) (dated 20 Jul 2026)
5. European Commission: Questions and answers on Article 50 of the AI Act (last updated 24 Jul 2026)
6. European Commission: AI Act policy page (checked 20 Oct 2026)
7. European Commission: AI Omnibus enters into force (published 27 Jul 2026)
8. European Commission: Commission publishes guidelines on transparency obligations (published 20 Jul 2026)
9. Gibson Dunn: EU AI Act Omnibus Agreement (published 27 May 2026)
10. Jones Walker: Yes, August 2 Still Matters (published 16 Jul 2026)