Rosen v. Pichai

What kind of case this is

Rosen does not sue as an author or a speaker. He sues as a stockholder, in Alphabet’s name, against the people who run it. The theory is that the directors and officers chose an “ask forgiveness not approval” approach to training data, that this exposed Alphabet to copyright and biometric privacy suits, and that they are personally answerable to the company for the cost (¶¶ 1, 8-14). Alphabet is named only as a nominal defendant (¶ 139).

The same plaintiff filed a comparable suit against Apple’s board, Rosen v. Cook, in August 2026. ChatGPT Is Eating the World counted that one as the fifth copyright derivative suit in the United States. This complaint adds biometric privacy, which is the stated basis for filing in Illinois: venue is pleaded on Alphabet’s Chicago presence and on voiceprints taken from Illinois residents (¶¶ 3, 17).

Three counts, and what sits beneath them

Copyright infringement and BIPA violations are not counts here. They are the alleged wrongdoing that the three counts are built on.

The underlying conduct alleged

• Text and images. Training on copies drawn from Z-Library, LAION-400M and LAION-5B (¶¶ 5, 60-70). For Z-Library, the complaint says the C4 dataset used for LaMDA and Gemini contains text from b-ok.org, a Z-Library domain, relying on a news analysis of C4 (¶ 70). The complaint points to the paper Google published for Imagen. That paper describes LAION-400M as “one of our data sources” and says the model was trained on about 460 million internal image-text pairs plus the roughly 400 million pair LAION dataset.
• Music. Sound recordings, compositions and lyrics taken from YouTube and Magenta to build MuLan, MusicLM and the Lyria models (¶¶ 5, 79-81).
• Voices. Hundreds of thousands of hours of speech used to pretrain Google’s commercial voice models without the written notice and release that 740 ILCS 14/15(b) requires (¶¶ 6-7, 90-106). The complaint contrasts this with Custom Voice and Chirp 3 Instant Custom Voice, which do require a recorded consent statement (¶ 90). Much of this section quotes the complaint in Marin v. Alphabet, No. 1:26-cv-05438 (N.D. Ill.) (¶ 106).

How it pleads that the board knew

Under Delaware law, Rosen must show that a demand on the board would have been futile. He needs to do that for five of the ten sitting directors (¶¶ 142-143, applying United Food & Commercial Workers Union v. Zuckerberg, 262 A.3d 1034 (Del. 2021)). The knowledge allegations come from public sources: the lawsuits already pending against Google, including In re Google Generative AI Copyright Litigation, No. 23-cv-03440-EKL (N.D. Cal.), and Kogon v. Google, No. 1:26-cv-02582 (N.D. Ill.). They also draw on earlier Illinois biometric settlements, suits against other AI developers, and the board’s own proxy statements saying AI is “regularly and extensively covered” at full board meetings (¶¶ 12-13, 109).

On damages, the complaint describes the harm to Alphabet as exposure to the underlying suits, including legal costs and lost goodwill, and lists executive and director pay for the period (¶¶ 134-136).