Copyright Law Update — November 29, 2023
By Matt Hersh

More rulings start to flow in on the intersection between artificial intelligence and copyright. In a decision just before the holidays, the Northern District of California brusquely dismissed portions of a class action brought by comedian Sarah Silverman and several other writers.

The lawsuit alleged that large language models, like those offered by OpenAI (the famed ChatGPT) and Meta, infringe their copyrights in several ways: by ingesting their works into their database (thus unlawfully copying the works) and by recasting their works as part of their output (thus unlawfully creating derivatives of their works). The lawsuits also brought claims under the DMCA and several common law theories.

The court easily batted away the derivative works theory – at least as applied against Meta – as “nonsensical.” To prevail on a theory that Meta’s outputs constitute derivative infringement, the court noted, “the plaintiffs would [] need to allege and ultimately prove that the outputs incorporate in some form a portion of the plaintiffs’ books.” The plaintiffs here did not do that, so their derivative works complaint failed. (The claim that Meta copied their works by hoovering them up into its system remains viable, at least for now.)

This is the second straight loss for a derivative works attack on large language models. In late October, Judge Orrick of the same court dismissed on similar grounds derivative works claims brought a series of artists against Stable Diffusion, the AI program that creates user-generated art.

I’ll have more to say about these issues in later posts.