What happened

CADA is a Sydney radio station owned by the Australian Radio Network (ARN), one of the largest commercial broadcasters in the country. For about six months, its weekday daytime slot ran a four-hour music show called "Workdays with Thy," hosted by a presenter named Thy. Thy played hip-hop, talked between tracks, and had a presence on the station's branding. Thy was not a person. The voice was AI-generated, cloned using ElevenLabs - a commercial AI voice-synthesis platform - from a real woman who worked in ARN's finance department.

Listeners were not told any of this. There was no on-air disclosure, no asterisk, no "this presenter is an AI" note. For months, a synthetic voice fronted a daily show on a major-market station, and the audience was left to assume, as audiences reasonably do, that the person talking to them between songs was a person.

How it unravelled

The show ran without incident until someone started asking obvious questions. In late April 2025, Stephanie Coombes, a Sydney-based writer, published a piece in her newsletter, The Carpet, poking at a simple gap in the record: who, exactly, is Thy? What is her surname? Where did she come from? A real radio host accumulates a public footprint - social media, prior bylines or air checks, a verifiable last name. Thy had a first name, a voice, and nothing else. The absence was the tell.

Coombes's questions forced the issue, and ARN confirmed what the silence had implied: Thy was an AI creation. The company said the voice and likeness were based on a real ARN employee from its finance team, that her actual photo and real voice had been used, and that she remained involved in the project. ARN did not, at least initially, clarify the terms of that involvement - whether and how the employee was compensated for having her voice cloned and broadcast under a fictional presenter identity for months.

The two-layer backlash

The reaction broke along two distinct lines, and they are worth keeping separate because they are different objections.

The first is the deception. A commercial station ran a synthetic host for roughly half a year without telling its audience. The problem is not that AI was used; experimentation with synthetic voices is a legitimate thing for a broadcaster to try. The problem is that it was concealed. Broadcast media runs on a baseline assumption that the human voice addressing you is human unless told otherwise. Quietly violating that assumption for months, and only confirming it when a writer refused to let the question drop, is the part that reads as a breach of faith rather than an experiment.

The second is the representation angle, which is what gave the story its particular sting. Thy presented as a young Asian-Australian woman. Asian-Australian presenters are underrepresented on mainstream radio, and an apparent diversity hire in a prominent daytime slot would normally be a small good-news story. Except the presenter did not exist. The "representation" was synthetic. Teresa Lim, vice-president of the Australian Association of Voice Actors, called the approach tokenistic, making the point that an apparent opportunity for representation on a major network had turned out to be completely fake - a manufactured face occupying a slot a real person could have held. As Lim put it, authenticity and truth matter for broadcast media, and the public deserves to know what the source is.

The Australian Association of Voice Actors pressed two concrete demands: a public apology, and mandatory labeling of AI-generated programming so that audiences are told when they are listening to a synthetic voice rather than a person.

What is and is not confirmed

A few things are worth stating plainly to avoid overclaiming. ARN's account is that the employee whose voice was cloned was a willing participant and remained involved. There is no public evidence that she was coerced, and the dispute over compensation is about transparency and terms, not an allegation of theft from her. The core, undisputed facts are these: the host was AI-generated; the voice was cloned via ElevenLabs from a real ARN finance employee; the show ran roughly six months; listeners were not told; and the disclosure came only after an outside writer publicly questioned Thy's identity in late April 2025.

The harm here is not a single dramatic incident but a sustained, low-grade deception plus the displacement effect. There is no claim that Thy gave dangerous advice or that anyone was financially defrauded. The damage is to trust and to the people - presenters and voice actors - whose jobs and visibility were quietly handed to a synthesizer.

Why it matters

This belongs to the same family as the Sports Illustrated fake-author scandal and the various newsroom episodes where AI-generated content was published under the cover of a human identity. The recurring failure is not the use of the technology. It is the decision to hide it.

The pattern is almost a template at this point. An outlet adopts a generative tool to cut costs or fill a slot. It dresses the output in a human identity - a byline, a headshot, a first name and a voice - so the audience never registers that anything has changed. The concealment works for a while because synthetic output is fluent and audiences are not auditing their radio hosts for proof of life. Then someone asks the obvious question the producers hoped no one would, the human identity evaporates on contact, and the outlet is left explaining why it ran a fake person as a real one.

Voice cloning sharpens the problem relative to text. A cloned voice carries warmth, familiarity, the sense of a relationship between host and listener that daytime radio is specifically built to create. Building that parasocial bond with a synthetic presenter, and not telling the audience, is exploiting precisely the trust the format depends on. The voice-actor association's demand - that AI programming simply be labeled - is not a radical one. It is the minimum a medium owes an audience that has historically been able to assume the person talking to them exists.

The fix is cheap and the industry already knows it: disclose. The reason these stories keep happening is that disclosure undercuts the trick. A synthetic host announced as synthetic is a novelty; a synthetic host passed off as a real diversity hire for six months is a scandal. CADA chose the second by default, and then had to explain it once a writer in Sydney declined to stop asking for Thy's last name.