Consider the following:

  • person A owns a lot of data
  • person B develops an algorithm that can be trained on this kind of data to produce a ML model that can make statistical predictions
  • person C develops a software that implements said algorithm
  • person D uses A's data, transforms it a little (cleaning, formatting, aggregations…) so that it fits the B's algorithm's expected input format, and trains it using C's software

I imagine it all depends on the license of the data / algorithm / software, but my question is who, via those kind of licenses (or in which conditions) has at least a partial claims on commercial benefits resulting in using the final model?

  • 1
    It might be useful to present this in terms of music recognition software, e.g. Shazam. Record company A owns music, university researcher B develops an algorithm, open-source developer C implements it, and finally a company like Shazam combines them. This may involve turning MP3 files into WAV files as MP3 is a compressed audio format unsuitable for ML.
    – MSalters
    Commented Jan 16, 2018 at 22:52
  • See also, "What IP law would apply to trained weights of an AI model?" answered here: law.stackexchange.com/a/90433
    – Cosmo
    Commented Nov 21, 2023 at 14:38

2 Answers 2


Person A has to have created the data to hold copyright; for most kinds of data this has no legal effect because facts are not protected by copyright. A mineable database probably does not have the necessary creative elements for copyright. An algorithm is not protected by copyright (it might be patented). Person C's program is copyrighted. The product created by D is probably copyrighted, depending on what degree of creativity is involved in their transformation. If the transformation is automatic then no, but if creative judgments are applied to the output of the program then maybe. Though the resulting product is another database of facts, and the facts cannot be protected.

In terms of "using the model", only C and possibly B have any control. If it is necessary to validate the software using A's data and A has kept the data secret, C might negotiate with A to use the data, in order to complete his program, and that could give A some interest in the program.


At this point in time, there appears to be not even a consensus on the question if a ML model can be copyrighted. Copyright protects creative works of various kinds, but it is not clear what type of work an ML model would be. Arguably, an ML model is produced by a non-creative automated process, which automatically excludes it from copyright protection. And if there is no copyright, then there is no copyright owner.

  • 1
    But then you do have some creative process in AI, especially in deep learning where the architecture of the neural network is designed by someone. But then again, you can argue it's an algorithm… Commented Feb 21, 2019 at 16:23

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .