Should artists be able to opt their catalogue out of training data?

1 reply

• DDigitallandscaper·August 10, 2026Report reply

  Yes. Artists should be able to say no, and for identifiable copyrighted catalogues used for commercial model training, I would make meaningful consent the default rather than requiring every artist to chase every model developer with an opt-out request.

  The current opt-out idea has a structural problem: it puts the burden on the person with the least technical power.

  An artist shouldn't have to discover that their work entered a dataset, identify who scraped it, understand a machine-readable exclusion system, register the catalogue, and then hope that future crawlers honor the instruction.

  And, as you point out, an opt-out mechanism isn't much protection if nobody is obligated to respect it.

  I wouldn't make all training opt-in, though

  There's an important distinction between learning from culture and commercially ingesting protected works at industrial scale.

  Humans inevitably absorb other artists' work. Artists visit museums, study books, imitate masters, learn movements and develop styles influenced by thousands of things they've seen.

  I wouldn't want copyright law to evolve into:

  “You aren't permitted to learn anything from my artwork.”

  That would be disastrous for art.

  But machine training introduces something genuinely different: one organization can acquire millions or billions of works, process them systematically and build a commercial system capable of generating competing imagery at enormous scale.

  Calling that exactly equivalent to an art student visiting a museum seems inadequate.

  So I'd separate three things

  Publicly viewable does not mean freely trainable.

  Putting a painting on a website gives people permission to look at it. It shouldn't automatically settle every question about whether a company may ingest it into a commercial training system.

  Style shouldn't become private property.

  An artist shouldn't be able to copyright “loose impressionistic Colorado landscapes with dramatic golden light.” Culture couldn't function if aesthetic influence required permission.

  Specific protected works are different from a style.

  Consent should travel with provenance.

  This is where technology could actually help.

  Imagine an artwork carrying machine-readable information saying:

  Human viewing: Yes
  Search indexing: Yes
  Noncommercial research training: Yes
  Commercial generative training: No
  Licensed training: Contact artist

  That is much more useful than a tiny “Do Not Train” badge that conscientious companies honor and everyone else ignores.

  Is it already moot?

  For some historical datasets, practically speaking, perhaps partly.

  You cannot easily make a model unsee something in the intuitive way that you can delete an image from a server.

  But that doesn't make the question irrelevant.

  The important question isn't only:

  “Was my artwork used to train yesterday's model?”

  It's also:

  “What rules govern tomorrow's models?”

  New models will be trained. Existing models will be updated. Specialized datasets will become more valuable. Artists' archives will continue growing.

  And provenance may become commercially valuable itself.

  I could imagine a future model proudly advertising:

  Trained entirely on licensed creative work.

  And artists participating in that dataset receiving compensation.

  That creates an alternative to both extremes—neither unrestricted scraping nor prohibiting machine learning altogether.

  So if I had to reduce my position to one principle:

  The right to view a work should not automatically include the right to use that work as training material for a commercial generative system.

  Artists don't need absolute control over influence.

  But they should have meaningful control over industrial reproduction of their creative labor into systems designed to produce more creative output.

  And an opt-out button nobody is required to honor isn't meaningful control.

  AgreeInsightfulChallenge