AI Act Transparency: Telling Applicants They Are Talking to AI

Short answer: yes, if the AI Act reaches you

If an admissions chatbot talks directly to applicants and the AI Act applies to your institution, applicants must be told they are dealing with an AI system, unless that is obvious. The transparency obligations in Article 50 have applied since 2 August 2026, so this is a live duty, not a future one.

The first question for a UK university, college or private provider is therefore whether the Act applies at all. The second is what a good notice looks like. This article covers both. It is general information, not legal advice.

Does the EU AI Act reach a UK institution?

The UK has no equivalent AI statute, but the EU regulation has extraterritorial reach. Under Article 2 of Regulation (EU) 2024/1689, providers and deployers established in a third country are in scope where the output of the AI system is used in the Union.

For admissions, the practical test is simple. If your chatbot answers questions from applicants located in Ireland, France, Germany or any other member state, or your vendor places it on the EU market, assume the Act is relevant and ask your counsel to confirm. A chatbot confined to UCAS-route applicants in Great Britain is a different case, but few institutions can say their audience stops at the border.

Even where the Act does not bite, UK GDPR transparency still does. The ICO expects people to be told clearly how their personal data is used, and a chatbot that collects names, emails and course interest is processing personal data from the first message.

What Article 50 says

Article 50(1) requires AI systems intended to interact directly with natural persons to be designed so that those people are informed they are interacting with an AI system, unless this is obvious to a reasonably well-informed, observant and circumspect person in context. Article 50(5) adds that the information must be clear and distinguishable, given at the latest at the first interaction, and accessible. The official text is on the Commission's AI Act Service Desk.

Three practical consequences follow:

• Timing: the notice appears in the first message, not on a terms page.
• Visibility: it must stand apart from the rest of the interface.
• Accessibility: it must work with screen readers and on mobile.

Provider or deployer: who carries the duty?

On the wording of the text, the Article 50(1) duty sits with the provider of the system, meaning the vendor who builds the chatbot. Your institution is normally the deployer.

That does not make it a vendor-only problem. You pick the tool, write the welcome message and answer to applicants, regulators and your governors. If you remove the vendor's notice, or build your own agent on top of a third-party model, you may be treated as a provider. Penalties are set in Article 99, and each member state designates its own authorities, so check who that is for the countries you recruit in.

What was not postponed

The Digital Omnibus on AI came into force on 27 July 2026 and did move some dates, but not this one. The Digital Omnibus moved Annex III high-risk use cases to 2 December 2027, not Article 50. The Commission's regulatory framework page sets out the timeline, and its transparency guidelines, published on 20 July 2026, restate the 2 August date.

The guidelines are aimed at authorities and are not a substitute for the regulation itself.

When you can skip the notice

The "obvious" exception is narrow. A chat widget titled "Virtual assistant" with a robot icon may be enough in some contexts, but your audience includes sixteen-year-olds, parents, mature students and international applicants with very different familiarity with chatbots.

Our advice is not to rely on it. One sentence is cheaper than arguing about what is obvious, and easier to evidence.

Watch four situations:

1. Handover: tell the applicant when a human joins, and when the AI resumes.
2. Other channels: on WhatsApp, SMS or voice, the notice belongs inside that channel.
3. Human-sounding personas: a named agent with a realistic photo suggests a person.
4. Automated emails: if an AI drafts a reply sent in your name, say so.

Sample wording

These are suggestions for your DPO to review, not wording prescribed by the regulation.

Avoid a bare "Assistant" label or a notice buried in terms and conditions. Both are hard to defend as clear and distinguishable.

AI notice and UK GDPR are separate duties

The Article 50 notice does not replace your privacy information. Applicants still need to know who controls their data, why, and for how long. Both can sit in the same welcome banner, with a link to your privacy notice.

For the wider picture, see our guide to the EU AI Act and higher education, the risk classification for schools and the documentation checklist.

Evidence to keep

A notice you cannot show is a notice you cannot defend. Keep:

• dated screenshots of each channel with the notice visible;
• the vendor contract clause covering user information;
• a change log for the welcome message;
• accessibility test results, including screen reader and mobile;
• the name of the person who approves wording changes.

Review the file whenever you add a channel, for example when a web-only chatbot gains WhatsApp.