Impact
EU AI Act Article 50: what to tell candidates
The only EU obligation that is live right now.
October 8, 2026
Ben Chino

The short answer
Article 50 of the EU AI Act requires that people are told when they are interacting with an AI system, unless it is obvious from the context. It has applied since 2 August 2026. In hiring it covers AI voice screening, AI-assisted interviews and conversational steps in an application flow.
Most of what has been written about the EU AI Act concerns obligations that are not yet in force.
In June 2026 the Digital Omnibus deferred the high-risk obligations, which include hiring and selection systems, from August 2026 to December 2027. Sixteen months. A lot of project plans quietly became wrong that week.
One thing did not move. Article 50 applies now, and has since 2 August 2026.
It is a much smaller obligation than the high-risk regime and it is the only one you have to meet today, which makes it worth understanding precisely rather than approximately.
What Article 50 says
Definition
Article 50
The transparency article of the EU AI Act. In the part relevant to hiring, it requires that people be informed when they are interacting with an AI system, unless that is obvious from the context.
It is not about fairness, audits or documentation. It is about the person knowing.
That is the whole requirement. Not fairness, not audits, not documentation. The person knowing.
Where it actually bites in a hiring process
Three places.
AI voice screening. A candidate receives a call, or makes one, and talks to a system rather than a recruiter. This is the clearest case and the one Article 50 was most obviously written for.
AI-assisted interviews. A live conversation where a system is listening, structuring or scoring. The candidate is talking to a human, so it is less obvious, which is exactly why the disclosure matters more.
Anything conversational in the application flow. Chat-based pre-screening, conversational forms, automated follow-ups that read as written by a person.
Note what is not on that list. A scored assessment that a candidate knows is an assessment is generally obvious from context. The cases that need the disclosure are the ones that feel like a conversation with a person.
The three mistakes
What fails the test
- A line on page four of the privacy policy
- "This process may involve automated decision-making technologies"
- English only, with the other five languages never checked
- A notice that appears after the interaction
What meets it
- The sentence appears where the interaction happens
- "You will be speaking with an AI interviewer"
- Every language you hire in, each approved by a speaker
- Before the interaction, with a record of where it is deployed
Burying it in the privacy policy. Informed means informed. A sentence on page four of a document nobody opens does not tell anyone anything. The disclosure belongs where the interaction happens, at the moment it happens.
Writing it in legal language. "This process may involve automated decision-making technologies" tells a candidate nothing. "You will be speaking with an AI interviewer" tells them something. The obligation is to inform, and informing is a test you fail by being unclear, not just by being silent.
Only doing it in one language. If you hire in six languages you need six versions, and someone has to approve each one. This is the step that gets found in an audit rather than in a review, because the English version is always fine.
What to actually do
The project
Five steps, finishable this month
Why this is worth doing properly even though it is small
The small project you can finish this month is also the first deliverable of the large project you have fourteen months to do.
Two reasons.
The first is that it is the only part of the EU regime you can be held to today, and it is cheap to meet. There is no reason to be exposed on it.
The second is less obvious. The December 2027 high-risk obligations will require documentation about systems, decisions and oversight. The exercise of listing every AI touchpoint in your hiring process, which Article 50 forces you to do now, is the first half of that work.
Key takeaways
Small obligation, live today, and widely missed.
- The EU AI Act high-risk obligations for hiring moved from August 2026 to December 2027. Article 50 did not move.
- It requires one thing: that people are told when they are interacting with an AI system.
- The three failures are burying it in the privacy policy, writing it in legal language, and only doing it in one language.
Frequently asked
Questions people ask about this
When did EU AI Act Article 50 take effect?
2 August 2026. Unlike the high-risk obligations for hiring, which moved to December 2027, Article 50 was not deferred.
Does Article 50 apply to a scored assessment?
Generally not, because a candidate taking an assessment already knows it is an assessment, which makes it obvious from context. It applies to the steps that feel like a conversation with a person: voice screening, AI-assisted interviews and conversational pre-screening.
Is a line in the privacy policy enough?
No. The obligation is to inform, and a sentence on page four of a document nobody opens does not inform anyone. The disclosure belongs where the interaction happens, at the moment it happens, in plain language.
Want to see how we word ours?
Including the translations. Ask and we will send them.
Ask for the wordingSources
EU AI Act Article 50 and the Council of the EU approval of the Digital Omnibus on AI, 29 June 2026. Last reviewed 8 October 2026.
See what Maki Agents can do for you



