When an AI Recruiter Has to Say What It Is

What Does the EU Rule Actually Require?

One rule in force speaks to the voice on the line rather than to what it does with your answers. From 2 August 2026, Article 50(1) of the EU AI Act requires providers to design systems intended to interact directly with people so the person is informed they are interacting with an AI, unless that is obvious in the circumstances ¹.

The duty falls on the company that built the screener, not on the employer that bought it.

That is worth knowing before you write to anyone. Your instinct will be to ask the employer why nobody said anything, and the design obligation in Article 50(1) does not sit with the employer at all ¹.

Read the obligation for what it covers, because it is narrower than the headline. It requires that you be informed you are interacting with an AI system. It does not require anyone to explain what the system does with your answers, to name who reads the transcript afterwards, or to offer you a person instead ¹. The exception for what is obvious to a reasonably well-informed and observant person is a judgment call, and it will be contested.

Being told is not the same as being told anything useful.

Why Don't the Illinois and California Rules Cover the Voice on the Line?

Because Illinois's video-interview law and California's coming notice rule both hang on what a system analyses or decides, not on whether it sounds human. Illinois has required since 2020 that an employer using AI to analyse a recorded video interview notify the applicant first, explain what the AI evaluates, and obtain consent ². A synthetic voice on a phone screen is not that.

The statute is narrow by design and narrower in effect. It covers AI analysis of applicant-submitted video interviews for Illinois positions, and nothing else: not resume screeners, not chat interviews, not a call. It also names no penalty and no private right of action, so the right to decline exists on paper before it exists in practice ².

A law with no remedy still tells you what the employer was supposed to do.

California's answer arrives later and asks for something else. Section 7220 of the state's finalised CCPA automated decisionmaking regulations covers a business using that technology for a significant decision. The business must give a Pre-use Notice at or before the point it collects the information it plans to process ³. The notice states the specific purpose in plain language, describes the right to opt out or, where a human appeal is offered instead, the right to appeal, and describes the right to access the technology. Compliance is required by January 1, 2027 ³.

The regulation's own example of a consolidated notice to a job applicant names resume-screening software and software that evaluates applicants' vocal intonation, facial expression and gestures ³. That is a rule about analysis again. It reaches the thing listening to you, not the thing speaking to you.

It also binds only businesses over the CCPA's revenue and data thresholds, which leaves many smaller employers out, and it protects only California consumers as the CCPA defines them, a group that includes job applicants in the state.

The same split shows up on the employer's side of the page. An AI hiring policy carries a candidate-notice section, and what belongs in an AI hiring policy sets out that section for the person drafting one. The notice duties it works from attach to a tool doing something, such as an automated employment decision tool screening a candidate. None of them fires because a machine introduced itself.

Ask, and Keep the Answer

At least one company building these interviewers has written answers for candidates in advance, and it costs you nothing to ask. Alex, the AI interviewer formerly named Apriora, publishes a candidate FAQ saying it makes no final hiring decisions and does not automatically reject anyone ⁵. Its recordings, transcripts and notes go to the employer's hiring team, where humans decide ⁵.

Alex is describing its own product there, which is not an audit of it. What the page does settle is that candidate questions about the process are ordinary. A company that has already written answers for candidates is not going to be startled when one asks about the process out loud.

Ask in writing if the answer matters.

Reply to the scheduling email and ask three things: whether the first round is conducted by an AI, whether it is recorded, and who reads the recording afterwards. Keep the reply. It is the only record you will have of what you were told and when, and it costs nothing to hold onto if the process turns strange later.

The format is common enough to plan for.

Alex's own About page claims 1,000,000 candidates interviewed and describes an always-on first-round interviewer that works by phone call, video or text, at a time the candidate picks ⁶. The counter is the vendor's, cumulative and unaudited, so treat it as evidence that the format exists at scale rather than as a measurement. The practical consequence is small and useful: a first interview can arrive as a phone call. Four kinds of AI interview, and what each one records sorts the formats apart before you are inside one.

What If Nobody Told You?

Not knowing is the common experience, not a personal lapse. In Greenhouse's 2026 candidate report, a survey of 1,200 US respondents, 70% said they were never clearly told upfront that AI would be evaluating them ⁴. Of the same 1,200, 21% said they found out only once the interview had started ⁴.

That is self-report from a company that sells hiring software, and it measures what candidates noticed rather than what a consent screen actually contained. Read it as a description of the experience. It is not a count of employers who broke a rule.

The gap it names is real either way.

In the same survey, 57% said disclosure should be a legal requirement, and 38% said they had already walked away from a hiring process because it included an AI interview ⁴. That last figure is a reported past action with no detail about whether the AI round was the only reason. Read it as a signal about how the format lands, not as a recommendation.

Where the entitlement already exists is a shorter list than where it should, and where you're entitled to know AI screened your application maps it. Neither of the two US rules above requires a company to say that the voice on the line is a machine; California's, from 2027, asks for notice of what the technology does with your answers, not of what it is. Asking is the move that works in every place, including the ones with no rule at all.

Read the evidence >