Short answer: if your store runs its own AI chatbot, negotiation bot, or conversational shopping assistant that talks directly to a shopper, EU law has required it to say so since August 2, 2026. That deadline didn’t move. A different, much heavier part of the AI Act got a widely-reported extension in July β€” and it’s easy to read the wrong headline and assume the whole Act got pushed back.

What Article 50 actually requires

Article 50(1) of Regulation (EU) 2024/1689 states it plainly: providers “shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect” (full Article 50 text). The obligation applies as of 2 August 2026, per the European Commission’s own guidance (Article 50 FAQ).

The Commission’s finalized guidelines set four cumulative criteria for when the duty kicks in: the system must qualify as an AI system; it must be built for a genuine two-way exchange with people, not just automated responses; the AI must communicate with the person directly, not through a human intermediary; and the person on the other end must be a natural person. Systems “operating solely in the background, through machine-to-machine communication, or without direct contact with people, fall outside the scope of this obligation.” The same FAQ names the covered category explicitly: chatbots, AI agents, and avatars.

What got pushed back β€” and what didn’t

The confusion comes from Regulation (EU) 2026/1744, the “Digital Omnibus on AI,” which entered into force 27 July 2026. It defers the AI Act’s heavy Annex III “high-risk” compliance regime β€” CE marking, EU database registration, a human-oversight override β€” from 2 August 2026 to 2 December 2027 for systems classified high-risk under Article 6(2) and Annex III, and to 2 August 2028 for Article 6(1)/Annex I systems (Regulation 2026/1744, Official Journal text).

That deferral is real, and it’s the one that would have swept in autonomous purchasing-decision systems. But it changes nothing about Article 50. The transparency duty was never part of Annex III, so it stayed on the original 2 August 2026 schedule β€” live for two and a half weeks now.

Where it applies to your store

  • Applies: an on-site AI chat widget, negotiation bot, or conversational shopping assistant that answers a shopper directly needs to disclose it’s AI β€” unless that’s already obvious from context.
  • Doesn’t apply: a third-party shopping agent (ChatGPT, Perplexity, Gemini) calling your ACP/UCP checkout endpoint or MCP tools on a shopper’s behalf is machine-to-machine β€” there’s no direct exchange with a natural person on your side of that call, so it falls outside Article 50(1)’s scope by the Commission’s own four-criteria test. Disclosing to the end user that they’re talking to an agent is that agent’s provider’s job, not yours.
  • The label only has to appear once, “at the latest at the time of the first interaction” (Article 50(5)) β€” it doesn’t need to repeat on every message.

Penalties

Enforcement sits mainly with national market surveillance authorities, and fines can reach €15 million or 3% of worldwide annual turnover, with proportionality considered for SMEs.

FAQ

Does my store need to label a shopping assistant if it’s obviously a bot anyway?

No β€” the Commission’s guidance builds in an “obvious” exception, assessed from the perspective of a reasonably well-informed, observant shopper given the context of use. But the guidance says this exception should be read narrowly, since it removes transparency from the person interacting with the system, so a generic “AI Assistant” label on a chat widget is the safer default rather than assuming obviousness will hold up.

Does this apply when ChatGPT or Perplexity buys from my store on a shopper’s behalf?

No. That interaction is between the shopping agent and the human user on the agent’s own platform β€” the disclosure duty falls on that agent’s provider (OpenAI, Perplexity, Google), not on the merchant whose checkout API it calls. Your store’s side of that transaction is machine-to-machine, which the Commission’s four-criteria test excludes from Article 50(1).

Did the EU AI Act’s August 2026 deadline get delayed?

Partly. The Digital Omnibus (Regulation 2026/1744) deferred the Annex III high-risk regime β€” the heavy documentation and registration rules that would have covered autonomous purchasing systems β€” to December 2027. Article 50’s transparency duty is a separate provision that was never part of that deferral, and it has applied since August 2, 2026.

Sources

A disclosure label is a trust surface, the same category of fact AgentReady’s audit already checks for agentic checkout β€” whether the machine-readable and human-readable parts of your store tell the truth about what’s on the other end of the interaction.