By the management of Radio Odyssey · 14 November 2026
Radio Odyssey plays tracks made with artificial intelligence. Not everywhere, and not on the quiet: in our "Discoveries" sequence, and flagged. This article is about the tracks of the artists we play; what we make ourselves is covered further down. It explains what we do, why, the rules we set ourselves, and above all what we cannot guarantee. We would rather you heard it from us.
What you hear on air
When a track is made with artificial intelligence, the mention (IA) — the French abbreviation for AI — follows the artist's name in the now-playing display. When the artist has a page in our Discoveries section, that page spells out how much of the track is AI: the music, the voice, sometimes both. It also says what we were able to check and what comes only from the artist or their label.
Two situations that should not be confused
Behind the same word, "AI", there are two very different realities.
The track generated by AI. It is produced largely by a tool, from instructions. It can be a "virtual" project, with nobody singing or playing: the artist is itself a creation.
The artist who relies on AI. They write, compose or arrange, then use AI as a tool, for the voice for example. The musical work comes from a person, upstream: the tool is new, the approach is not. It is not the same thing, and we do not want to lump them together.
In both cases we flag the AI with the mention (IA). What differs is what the artist's page says about it.
At the very least, we expect the lyrics to be written by a person. We cannot check this ourselves: we rely on the good faith of the artist and the label.
In July 2026 the BBC took a stricter position than ours: it gives priority to music that comes from meaningful human creativity, and considers that simply prompting, generating, selecting or lightly modifying an AI output is probably not enough. We do not set the bar in the same place: we also accept fully generated projects, provided they are presented as such and that there is a real artist, label or artistic project behind them.
Our rules
- We flag it. For the tracks of the artists we play: (IA) in the now-playing display, details on the artist's page.
- There must be someone real behind it: a real artist, a real label, or a real artistic project.
- No imitation, no copying, no identity theft. We do not accept a track that imitates the voice of an existing artist, copies an existing work or passes itself off as a person. An AI track is presented as an AI track.
- In Discoveries, and nowhere else. Artists' tracks made with AI belong in the "Discoveries" sequence. Outside it, we do not play any, except a track that breaks through in the charts and that the public has embraced.
- Rights. For a track submitted by an artist or a label, we rely on their declaration about the origin of the track and the rights attached to it. We cross-check what we can, without being able to verify everything: the declaration binds the person who makes it. We aim to respect the law in force for as long as a track stays on air.
- If a declaration is inaccurate, we remove. If a serious element shows that a track was submitted to us without the use of AI being disclosed, we suspend its broadcast and question the artist or the label. If the omission is confirmed, we remove the track from the air, along with its page on the site.
Why we play it
Because you cannot go against a trend like this, and it is going to grow. There are already web radios whose music is entirely generated by AI: IA Radio, which presents 25 stations playing only original creations, or DaizaRadio IA, which announces only original compositions generated by AI. Some go as far as handing the microphone to virtual presenters: that is what RadioGPT, developed by the American company Futuri, offered as early as 2023. Pretending this does not exist would help nobody.
There is also a more positive reason. For emerging labels and artists, AI is today a real opportunity to create and be heard, without the studio, the musicians and the budget that used to be required. The internet brought out podcasters and unknown talents who had no way of breaking through in the big media. We think something comparable is happening here, and our "Discoveries" section exists precisely to give a chance to artists you do not hear elsewhere.
We are taking this trend head-on, and we are thinking it through. Radio Odyssey is not a debate station and we cannot open the discussion on air. But we stay attentive, and we can turn back at any time if we see excesses that go too far.
We can be fooled
Let us be clear: we cannot guarantee that no track was made with AI without our knowing. The human ear is no longer enough. In a survey commissioned by Deezer and run by Ipsos among 9,000 people in eight countries, 97% of respondents could not tell, in a blind test, fully AI-generated songs from a human-made one.
Detectors exist, but none is infallible. As of the date of this article, Deezer uses a tool that spots traces left by certain generators, inaudible to the ear. YouTube Music relies both on its partners' declarations and on its own detection tools. Apple Music, for its part, has introduced labels that record labels and distributors must fill in themselves, with no detection by Apple. In other words, labelling rests largely on what each party declares.
We have already had to set aside a track whose origin looked doubtful: an AI-generated version of an upcoming track, released before the official release, which closely resembled the real version published shortly after. These cases will multiply.
This is how we work:
- we rely on what the artist or label tells us;
- we cross-check what we can (the track's presence on streaming platforms, its release date, its audience, the press);
- on our pages, we write "according to their press kit" for what we could not verify.
When a track carries no (IA) mention, it means no AI was reported to us. It is not a guarantee, and we do not want to present it as one. If an error is brought to our attention, we look into it: if it is confirmed, we remove the track and its page, or we correct the mention.
What others do
Here is, as of the date of this article, what others do. The movement is general, and the choices differ.
- The BBC published its policy on AI in music in July 2026. It gives priority to human creativity, accepts that artists use AI tools as long as they do not replace their creative work, requires artists and partners who submit tracks to say whether and how they used AI, and promises never to knowingly broadcast generated music that would infringe copyright.
- YouTube Music began on 14 October 2026 to display a label on tracks created primarily with generative AI, for instance when the composition and vocals are generated. The label appears on the player page and opens an explanation.
- Deezer detects, labels and removes from recommendations tracks that are fully generated by AI, which are also kept out of its editorial playlists. According to its press release of 21 July 2026, these tracks made up more than half of the new tracks received each day at the June 2026 peak, around 90,000 tracks a day, for only 1 to 3% of listening. Deezer also states that, according to its analyses, up to 85% of the streams of these tracks were fraudulent in 2025. That is one of the reasons we ask for someone real behind every track.
- Apple Music introduced transparency tags in March 2026 (artwork, recording, composition, video), filled in by labels and distributors.
- iHeartMedia, a major American radio group, launched its "Guaranteed Human" approach at the end of 2025: no AI-generated voices on air, and no music with synthetic singers pretending to be human.
Some therefore make a very different choice from ours, notably by keeping these tracks out of their editorial programming. We respect those choices. Ours is to play them, saying so, within a limited framework.
And the law?
The legal framework is taking shape. The European regulation on artificial intelligence contains, in its Article 50, transparency obligations that have applied since 2 August 2026. Providers of tools that generate sound, images or text must mark the outputs in a machine-readable way; a grace period nevertheless runs until 2 December 2026 for tools already on the market before 2 August. Those who use an AI tool to generate or manipulate "deepfake" content (image, sound or video that resembles real people, places or events and could falsely appear authentic) must disclose that it is generated; for an artistic or creative work, this disclosure must be made in a way that does not hamper the display or enjoyment of the work.
We are not lawyers and we do not claim to say how these rules apply to a web radio. All of this is moving fast: we follow these developments, we will adapt our practices accordingly, and this article will be updated. Until then, flagging AI, as we do, is the common-sense rule that several major players in the sector have adopted.
What we produce ourselves
Part of what the station makes itself is produced with the help of artificial intelligence tools: all the sung jingles, voices and music, and all the music of the heart coherence breathing sequences. The sung voices of these jingles are synthetic; the voices of the breathing sequences and of the other jingles are human. These elements are not announced one by one on air; the information is given here and in our legal notice.
Behind all of this, humans decide. All editorial decisions, programming, choice of tracks and published content, are made by the management of Radio Odyssey, which takes responsibility for them. Artificial intelligence tools assist; they do not decide.
What this changes for you
The radio is a single stream: we cannot offer you a version without AI. What we can do is read the messages you send us about this, and take them into account. You can write to us at the address shown at the bottom of every page, to flag a track you think is wrongly labelled, or simply to tell us what you think.
Important note
Last updated: 14 November 2026.
Radio Odyssey chooses the tracks it plays and takes editorial responsibility for them.
Information about the origin of a track, notably any use of artificial intelligence, and about the rights attached to it, is declared to us by artists, labels or their representatives. These declarations bind those who make them. We cross-check what we can, without being able to verify everything. The absence of an (IA) mention means that no AI was reported to us, not that there is none.
This article describes our practices as of the date shown above. They may change with the law and with the technology.
If you believe that a track we have broadcast infringes your rights, your voice or your image, or that a mention is inaccurate, write to us at odysseylaradio@gmail.com stating the track concerned and the reason. We reply to every specific request. When it is serious, we suspend the broadcast of the track while we examine it. If it is well founded, we remove the track and its page, or we correct the mention.
Frequently Asked Questions
Sources and references
- BBC — "The BBC's policy on AI in music" (Lorna Clarke, July 2026)
- YouTube — "YouTube Music adds new labels for AI music" (14 October 2026)
- Deezer — "AI music has surpassed 50% of new music uploads for the first time" (21 July 2026, including the Ipsos survey of November 2025)
- Regulation (EU) 2024/1689 on artificial intelligence — Article 50, transparency obligations (European Commission, AI Act Service Desk)
- ActuIA — AI Act timeline after the omnibus, Regulation (EU) 2026/1744, in French (updated 6 October 2026)
- Apple Music transparency tags (March 2026) — Music Business Worldwide
- Apple Music transparency tags (4 March 2026) — 9to5Mac
- iHeartMedia, "Guaranteed Human" (November 2025) — RadioInsight
- IA Radio (official website)
- DaizaRadio IA (official website)
- RadioGPT, a Futuri station — Journal du Geek, in French (3 March 2023)
Sources consulted on 14 November 2026. Information about Apple Music and iHeartMedia comes from the trade press. In case of any difference between this translation and the French version, the French version prevails.
