The AI agent that doesn't introduce itself as AI is now illegal in Europe — and here's what that changes

Two things happened in the first week of August that show where the agent market stands. On July 29, Microsoft reported that its Agent 365 ecosystem accumulated 40 million registered agents in just two months since launch — for the first time exceeding 30 million human Copilot seats sold: in the world's largest enterprise ecosystem, agents now outnumber people. Three days later, on August 2, the European Union made mandatory something most companies hadn't put on their checklist: every agent that speaks with a person must identify itself as AI before its first response. Not in the terms of service — in the conversation itself.

What Article 50 actually requires — in plain business language

Article 50(1) of the EU AI Regulation (Regulation 2024/1689) applies to any AI system that interacts directly with people: chatbots, service agents, voice assistants, avatars. The obligation for the provider or deployer is concrete: ensure the user knows they are interacting with an AI. The notification must occur at or before the first interaction. What Article 50 rules out as sufficient compliance: a mention buried in terms and conditions, a vague reference to a “virtual assistant,” a footer text. The notification must be perceivable within the conversation channel — without the user having to look for it. And if a company chains multiple agents — one for qualification, one for service, one for escalation — the obligation applies to each node that interacts directly with a person, not just the first.

Why an agent that introduces itself converts better than one that doesn't

Article 50 compliance is a legal floor, not a ceiling. But there's a business case independent of the regulation: users who know from the start they're speaking with an AI and choose to continue have higher resolution rates and lower abandonment than those who discover it mid-conversation. The dynamic isn't paradoxical: whoever continues the interaction knowing it's with an agent does so with aligned expectations. The surprise of discovering it late — when the agent can't do something a human would — is what destroys trust. An agent with its own name, a clear role, and explicit AI disclosure converts better precisely because it doesn't pretend to be something it's not.

• Agent's own name: not “Virtual Assistant,” but a name with a role — “I'm Luz, commercial director at InnovaBlack Agents” — meets Article 50 and anchors the agent's identity from turn one.
• Automatic first message: the agent opens the conversation with a phrase that includes its AI identity, before the user types anything — the cleanest standard for satisfying “at or before the first interaction.”
• In agent chains: if the first agent hands off the conversation to another, the receiving agent must also introduce itself — the obligation doesn't transfer: it replicates.
• If you operate outside the EU: Article 50 applies to any company making agents available to users in the EU, regardless of where the company is registered — the European standard is already the global standard for agents with international reach.