IP Hot Topic: Vicarious Liability and AI: Alphabet’s Exit Highlights Corporate Shields

Last week, the Northern District of California untethered Alphabet from Google in a copyright infringement action, while Google will continue to face certain infringement claims. The court made clear that ordinary parent–subsidiary ties do not create vicarious liability—plaintiffs must show a parent’s practical ability to detect or stop infringement. The ruling underscores a bright line in artificial intelligence disputes: corporate oversight alone will not pull parent companies into the fight.

As factual background, this case concerns allegations by several authors and visual artists that Google and its parent Alphabet infringed their copyrights by using their works in training datasets (e.g., C4, Infiniset, LAION-400M, LAION-5B) to develop Google’s generative AI models, including PaLM, LaMDA, Bard, Gemini, and Imagen. Plaintiffs asserted claims of direct infringement against Google and vicarious infringement against Alphabet. While the claims against Google focused on the use of copyrighted works in training datasets, the claims against Alphabet rested on its role as the parent company with alleged oversight of Google’s AI operations

Vicarious liability for direct copyright infringement requires both (1) the legal right and practical ability to supervise and limit the infringing conduct as well as (2) a direct financial interest in the infringing activity. Finding inspiration in cases where the defendant exerted control over a physical infringement locus, the 9^th Circuit has applied the same standard (i.e., the right to control the space and the ability to police and stop conduct in that space) to operators of online platforms, services, and websites. The court observed, “The defendant’s vicarious liability often turned on the technological feasibility of detecting and stopping direct infringement by third-party users of the operator’s platform or service.” For example, in Napster, the relevant files were named by users and could not be accessed by the defendant to determine their true content.

Here, the court found that Plaintiffs failed to plausibly allege that Alphabet had the practical ability to limit Google’s direct infringement. Plaintiffs alleged only that Alphabet had “control authority” as parent company and could implement “copyright compliance measures.” But no facts showed Alphabet’s involvement in dataset selection or training and no allegations of Alphabet’s technical ability to detect/stop copying. Courts in analogous digital cases (Napster, Perfect 10 v. Amazon, VHT v. Zillow) required evidence of technological feasibility to police infringement; plaintiffs offered none.

Second, the court found no basis in the Copyright Act or case law to hold a parent company vicariously liable solely due to the parent-subsidiary relationship. The court rejected Plaintiffs’ argument that Alphabet, as the parent company of Google, was vicariously liable because it failed to use its “control authority” to stop the infringing conduct. Parent companies are not automatically liable for subsidiaries’ acts absent alter ego or direct participation. Plaintiffs alleged only ordinary parent-subsidiary overlap (shared executives, strategic oversight, common financial incentives), which are insufficient as a matter of law. Allegations that Alphabet set “strategic AI direction” were typical high-level oversight, not control over infringing acts. Plaintiffs did not allege that Alphabet had any direct or indirect role in Google’s training of its models. Without more, accepting Plaintiffs’ theory would create automatic parent liability for any subsidiary infringement, contrary to corporate law principles.

The court thus dismissed Plaintiffs’ vicarious liability claims against Alphabet with prejudice.

Conclusion & Takeaways:

The court’s dismissal of all claims against Alphabet underscores that ordinary parent–subsidiary oversight is not enough to establish vicarious liability in AI copyright cases. Plaintiffs must plead concrete facts showing a parent’s practical ability to detect or stop infringement, not just strategic alignment or shared executives.

For AI companies, the key takeaways are:

• Corporate separateness can be a strong defense. Plaintiffs cannot bootstrap vicarious liability against deep-pocketed parents without concrete facts.
• Allegations that a parent set approval gates for training data, managed compliance processes, or directly participated in dataset selection could pierce this corporate shield.
• While oversight does not mean liability, operational entanglement could, so governance should be documented carefully.

For copyright holders, the key takeaways are:

• Target the right entity. Courts will not assume parent companies are liable just because they own or oversee a subsidiary—claims must be aimed at the operating company that actually trained or deployed the model.
• Allege concrete control, not just oversight. To reach a parent, plaintiffs must plead facts showing it had a practical ability to detect or stop the infringing activity (e.g., approval authority over datasets, direct role in training governance).
• Plead with granularity. General references to corporate structure, strategic direction, or shared leadership will not suffice; successful claims will need detailed allegations tied to specific datasets, models, and decision-makers.