90 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 13 July 2026
← The incident library
HD-INC-061
Media · Australia · 2025 · Undisclosed AI

A Sydney station ran an AI-generated radio host named 'Thy' for six months without telling listeners she wasn't real

By Ellie Harris · Filed On air for ~6 months; revealed April 2025 (Sydney, DAB+ and iHeartRadio)

Alleged: Australian Radio Network (ARN); ElevenLabs developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

A Sydney station ran an AI-generated radio host named 'Thy' for six months without telling listeners she wasn't real

What happened

CADA, a Sydney station owned by the Australian Radio Network (ARN) and carried on DAB+ digital radio and the iHeartRadio app, ran a weekday music show, “Workdays with Thy.” For approximately six months before it was revealed in April 2025, Thy introduced tracks and filled the space between songs in a warm, confident, young-sounding voice. That voice was not a person. It was reportedly generated with the AI voice company ElevenLabs and modelled on a real ARN employee, and it was presented to the audience as a host named Thy, accompanied by promotional imagery depicting the presenter. Listeners were reportedly never told she was synthetic.

The gap was noticed in April 2025 when the Sydney writer Stephanie Coombes, in her newsletter The Carpet, asked the obvious question no one could answer: who is Thy? There was no surname, no biography, no history anywhere. ARN then confirmed the host was AI-generated. At the time there were no specific broadcasting rules against an undisclosed AI presenter, so the station had breached no code; the objection was to the silence. Teresa Lim, of the Australian Association of Voice Actors, reportedly called the move deceptive, and noted the added sting that a synthetic persona was reportedly styled as a young Asian woman while real voice artists went unhired. ARN reportedly framed the exercise as a trial that had delivered valuable insights and said it reinforced “the power of real personalities.” In the aftermath the rules changed: under the Commercial Radio Code of Practice 2026, the ACMA now requires stations to disclose when a synthetic voice hosts a regularly scheduled program, effective 1 July 2026.

What an auditable version would have shown

The question that took months and a journalist to answer, whether the host is a person or a program, should be answerable in a single line at any moment. An auditable version binds every broadcast segment to a record of what produced it: that the audio was machine-generated, which voice model and version made it, whose voice it was cloned from and with what consent, and when it aired. Disclosure then stops being a favour the broadcaster may withhold and becomes a field in the record, and the same entry that would tell a regulator the segment was synthetic can drive the on-air and in-app label that tells a listener the same thing.

Where the gap was

The failure wasn’t that AI was used. The failure was that listeners were never given accurate provenance about what they were hearing. The host’s identity, that Thy is an AI voice and not a presenter, was not attached to the thing it produced; it lived in a marketing decision that could choose to say nothing, rather than in the output, which would carry its own provenance. PersonaGuard would enforce that a system’s presented identity matches its actual identity, preventing a synthetic presenter from being represented as a human without explicit disclosure. ConductRecord would preserve an immutable record of how each segment was generated: the model, the source voice, consent, timestamps and disclosure metadata, so the fact of AI is recorded at the source, not admitted after someone forces the question.

What governance should have looked like

“Is this a real person?” is among the first things an audience is entitled to know, and the answer should be built into the broadcast rather than supplied on demand. A synthetic presenter is a defensible editorial choice; running one styled as a human, and only explaining the choice once questioned, is where trust broke down. Provenance recorded at the moment of generation, a persona that cannot misrepresent itself, and a disclosure that follows the audio wherever it plays would have made the six-month mystery a non-question, and left ARN able to show exactly what it did and when, instead of confirming it after the fact. Trust wasn’t lost because AI hosted the show. Trust was lost because listeners couldn’t distinguish between a human voice and a synthetic one.

The reference implementation of PersonaGuard and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and Australian Radio Network (ARN); ElevenLabs could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

Last reviewed June 2026. This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →