UPDATED on Nov. 7, 2025: We posted an updated graphic compiling all the U.S. copyright lawsuits against AI companies in which the authors are advancing the Shadow Library Strategy against the AI companies.
The Shadow Library Strategy is to raise a separate theory of infringement–apart from the training of AI models–based on the AI company’s initial acquisition of copies of works from controversial shadow libraries.
It is still a developing theory–having been accepted by one judge only for the purposes of rejecting a fair use defense on summary judgment while another judge rejected the theory:
(1) Judge Alsup accepted this theory in rejecting Anthropic’s fair use defense on summary judgment as to “library building” copies in Bartz. (Judge Alsup later explained that Anthropic could re-assert fair use at trial before the jury, but the case settled.)
(2) But, in Kadrey, Judge Chhabria rejected this theory, ruling that the initial acquisition of copies was for the further purpose to train Meta’s model, a transformative use.
As you can see from the graphic below, this Shadow Library Strategy is being asserted against nearly all of the major US AI companies. The only ones are not on this list are Google (the two lawsuits against it do not allege use of shadow libraries presumably because Google has its own Google Books database), as well as (1) Amazon and (2) Elon Musk’s xAI. Neither has been sued.
Why is the Shadow Library Strategy popular?
The simple reason is that the Shadow Library Strategy has worked in 1 case so far, Bartz v. Anthropic, resulting in the largest copyright settlement in US history ($1.5 billion).
Plus, for the purposes of statutory damages, the plaintiffs would recover nothing more ($0) if they prevailed also on the AI training issue. Statutory damages is calculated per work infringed, no matter how many copies of the same work was made. For example, a defendant who downloaded a copy of 1 work and made numerous copies of the same work to train an AI model could only recover for 1 work infringed, assuming fair use were rejected across the board.
Related Stories