Can you claim copyright over a social media post that is used to train an AI model?

Question

Can user-generated content from social media platforms be considered copyrighted training data for AI models, despite agreements between the platforms and AI companies (note that the users themselves have no agreement with the AI company)?

What if an AI model's output is very similar to the training data ingested from this source. Can that output be considered copyrighted?

Are there first-principle arguments for the same, and is there some part of the technical process of training a model that can help navigate this?

What about the argument that if the model has spat out something very similar to the source it has been trained on, the original source did not have much (or any) creative content to begin with, therefore not allowing for copyright in the first place?

Answer 1

There is no general answer to this question, but usually application of copyright law won't help by itself.

Many social media sites, especially ones with free access, have rights to reuse your data, that you will have agreed to in order to use the service. Those will generally be written so that any agreements the media companies then make onwards to sell or give away your generated content are valid. However, some more general rights like those addressed in Europe's GDPR might conflict for some kinds of data.

Whether you have any legal standing or rights over content you posted is going to depend on the specific social media service, the content, and the AI service in question. If you have those details, then you may be able to ask a more concrete question. If there's anything serious at stake you will need to involve a lawyer. If you want to get a more open opinion on your rights and options for a specific case, you could ask elsewhere, e.g. on Law Stack Exchange.

There are some grey and unresolved legal areas around non-consensual data harvesting used to train generative models. Some large cases may be decided one way or another (which you could use to approximately assess your own cases if they are similar), and there is also the possibility of new laws that give more precise answers in future.