Putting an AI Agent in Front of a Prospect: What Article 50 Says

I run sales at Corefy, so the question of where an AI agent is allowed to talk to a prospect on its own is a practical one for me. In the EU it now has a legal angle too. Article 50 of the AI Act has applied since 2 August 2026, and the Digital Omnibus did not postpone it. I read the article as published on artificialintelligenceact.eu, which carries the Omnibus amendments. I’m not a lawyer, and this is about what the text says and where I’d be careful, not legal advice.

What Article 50 says that matters for sales

Article 50(1) requires providers to design AI systems that interact directly with people so that those people are informed they’re dealing with an AI. The exception is where it is obvious to a “reasonably well-informed, observant and circumspect” person, given the circumstances. The test is objective, so what the vendor believes is obvious doesn’t count. Article 50(5) adds that the information has to be clear and given at the latest at the first interaction.

Two things follow. The duty in 50(1) falls on the provider, the party that builds the system. A sales team that buys an AI chat or email agent is usually the deployer, so the first question for the vendor is whether the product tells people what it is, by default and in a way you can’t switch off. A company that builds its own agent may be the provider and carry the duty directly.

Article 50(4) is the one most summaries stress, and I think it’s less relevant to outreach than it sounds. For text, the disclosure duty applies to AI-generated text published to inform the public on matters of public interest, and it doesn’t apply where a person has reviewed the text and holds editorial responsibility. A one-to-one sales email isn’t published to inform the public, so on its face I don’t read 50(4) as covering it.

Where I’d be careful anyway

I can’t tell from the text whether a drafted email counts as an AI system “interacting directly” with someone when a human presses send. I’d guess not, but that is a guess. An autonomous agent that holds a back-and-forth conversation with a prospect, over chat, email or voice, looks like a clearer case, and an agent that poses as a named person looks like the opposite of what the article is trying to prevent.

The practical rule I’d use is simple: if the agent can reply without a human reading the reply first, tell the person it’s an AI in the first message. It costs a line of text and removes the argument about whether it was obvious.

I haven’t looked at data protection, e-privacy or national marketing rules, which can apply to automated outreach regardless of the AI Act, and I’d check those before this.

Keeping a person in the loop

Separate from the law, there is a design question about which actions an agent may take alone. The model I find most convincing is the one in Visa’s Agent Pilot Pays Only What a Human Approved, where the agent acts but a person sets the limit it acts within. In sales that could mean agents draft, research and schedule freely, while anything that makes a commitment, such as pricing, terms or a promise about delivery, needs a human sign-off.

Article 4, on AI literacy, has applied since February 2025 and sits next to all of this. In the amended text, providers and deployers are to take measures to support literacy among staff using AI on their behalf, with no specific level required. For a sales team, I’d read that as: whoever sends from an agent should know what it can get wrong. I wrote about the dates and what didn’t move in The EU AI Act’s High-Risk Dates Moved. Two Obligations Didn’t.