Beggars Group told AI companies no in writing — here's exactly what the notice says

Most of the AI-and-music fight plays out in courtrooms, where it is slow and expensive. Beggars Group has taken the cheaper route: publishing a standing legal notice on its own website and daring anyone to claim they did not see it.

The notice sits at beggars.com/reservation-of-rights-ai↗ (opens in a new tab). It is a few paragraphs long. It covers the catalogues of XL Recordings, 4AD, Matador, Rough Trade and Young, among others. And it is written to be as wide as the law allows.

What the notice actually prohibits

The operative sentence is deliberately exhausting. Beggars prohibits "any past, present or future access, use, copying, reproduction, extraction, distribution or exploitation" of its works "for the purpose of text and data mining, Artificial Intelligence training, development or commercialization of AI systems, tools or technology, web scraping, mining, altering, making extractions, preparing derivative works or similar purposes, and by any means."

Three things in that clause do a lot of work:

• "Past, present or future." It is retroactive by intent. A model trained in 2023 is not grandfathered in by this reading.
• "Or portion thereof." Snippets count. A thirty-second clip scraped from a preview player is not a loophole.
• The content list. It is not just masters. The notice names "recordings, audiovisual recordings, composition, lyrics, artwork, images, data, metadata, etc." Metadata being in that list matters — a lot of scraping projects treat credits and tags as free public data.

There is a separate sentence for the thing artists actually worry about: "The use of any artist name, image, likeness and voice also requires specific permission." That is a voice-clone clause in fifteen words.

The EU part is the interesting part

The notice expressly reserves rights "under Article 4 of the EU Directive 2019/790/EC." Article 4 is the text-and-data-mining exception in EU copyright law. It permits mining of lawfully accessible works unless the rightsholder has reserved their rights in an appropriate machine-readable manner. Publishing a reservation is how you switch the exception off for your catalogue.

Then Beggars adds a parenthetical that reads like it was fought over by lawyers: they reserve under Article 4 "to the fullest extent permitted... (in the event that a court decides this Article might apply to the above uses, an interpretation we should add we do not agree with)."

Translated: we do not accept that the TDM exception ever covered training a commercial music model in the first place, but if a court disagrees, we have opted out anyway. Belt and braces. It costs nothing to say and removes the "they never reserved" defence.

The reservation also applies to content "identified through publicly available means" — aimed squarely at the argument that if a track is on a public streaming page, it is fair game to crawl.

Why a notice instead of a lawsuit

Litigation is the loud option. Wixen's publishing claim against Meta is the loud option. A published reservation is the quiet one, and it does something a lawsuit cannot: it fixes the date. Once the notice is public and machine-readable, an AI developer arguing good faith has a much harder story to tell about anything scraped afterwards.

It also scales. Beggars does not need to know who is crawling. The notice applies to everyone equally, forever, without a single filing fee.

The limitation is just as real. A reservation is only as strong as the enforcement behind it, and enforcement still means going to court eventually. What the notice buys is a better starting position when that day comes.

What this means if you own recordings

You do not need a label's legal budget to copy the structure. The useful pieces:

1. Put it somewhere permanent and crawlable — a page on your own domain, linked in your site footer, not buried in a PDF.
2. Name every asset class, not just the audio: artwork, lyrics, metadata, video, press photos.
3. Reserve explicitly under Article 4 if any of your catalogue is exploited in the EU, and say you are doing it without conceding the exception applies.
4. Add the name/image/likeness/voice sentence. Copyright in a recording does not automatically cover a voice clone; that is a separate right in most places and it needs its own line.
5. Back it with machine-readable signals — robots.txt disallow rules and a noai directive are weaker than a legal notice but they are the thing a crawler actually reads.

None of this stops a determined scraper. It does change what happens afterwards, which for now is the part rightsholders can control.