A federal judge has significantly narrowed the scope of a class action lawsuit against AI companies accused of using artists' works without authorization to train their AI model, Stable Diffusion. In the case, Andersen et al. v. Stability AI Ltd., artists led by Sarah Anderson alleged that the defendants, including Stability AI Ltd., Stability AI, Inc., DeviantArt, Inc., and Midjourney, Inc., infringed their copyrights by training their AI on billions of images scraped from the internet, including the plaintiffs' works.

Background of the Case:

• Plaintiffs' Allegations: Artists led by Sarah Anderson filed a class action lawsuit against AI platforms, including Stability AI Ltd., DeviantArt, Inc., and Midjourney, Inc., alleging infringement of their artwork through the AI software product Stable Diffusion.
• Claims Presented: The plaintiffs presented multiple claims, including direct and vicarious copyright infringement, violations of the DMCA, Right to Publicity under California law, and Unfair Competition.

The court dismissed most of the claims due to the plaintiffs' failure to register their works with the Copyright Office, a prerequisite for filing an infringement suit. However, the court allowed the copyright infringement claim to proceed for registered works, acknowledging the plausibility that they were included in the training datasets.

Court's Decision:

• Motions to Dismiss: The court granted the defendants' motions to dismiss, based on the plaintiffs' failure to register their works with the US Copyright Office, a prerequisite for filing an infringement suit.
• Specificity in Allegations: The court required plaintiff Anderson to specifically identify her works used as training images for the AI software, Stable Diffusion, noting that a search of her name on "ihavebeentrained.com" supported her claims' plausibility.
• Uncertainty about AI Functioning: The court acknowledged the lack of clarity on how exactly the AI platforms utilize the copyrighted training images and their liability for direct infringement.

The case highlights the legal complexities emerging around AI and copyright law, specifically regarding derivative works and the extent to which AI-generated images can be considered infringing. The judge expressed skepticism over the plaintiffs' argument that output images from Stable Diffusion, which are generated in response to text prompts and are derivatives of the training images, are substantially similar to specific copyrighted works.

Key Observations:

• Derivative Works Debate: The plaintiffs argued that the output images from the AI software are derivative works of the training images, a theory the court found lacking evidence of substantial similarity.
• Skepticism Over Direct Proof of Copying: The judge expressed skepticism over the plaintiffs' argument that direct proof of copying negates the need to show substantial similarity between the original and the derivative works.

Implications:

• Rising AI-Related Legal Cases: This case is among the many AI-related legal battles emerging, setting a precedent for how similar cases might be handled in the future.
• Importance of Copyright Registration: The ruling underscores the necessity for artists to register their works with the Copyright Office before pursuing infringement litigation.

This ruling is not yet an appellate court decision but represents an important line of analysis in the evolving legal landscape surrounding AI and copyright infringement.