---
id: "concept-fair-use-divergence"
type: "concept"
source_timestamps: ["¶4", "¶5", "¶6"]
tags: ["copyright-law", "fair-use", "litigation"]
related: ["entity-judge-william-alsup", "entity-judge-vincent-chhabria", "quote-alsup-transformative", "quote-chhabria-competing", "prereq-fair-use-doctrine", "question-appellate-resolution"]
definition: "The current split in federal courts over whether training LLMs on copyrighted material is transformative fair use or market-supplanting infringement."
sources: ["tail2"]
sourceVaultSlug: "hbr-seg-tail2"
originDay: 2
articleStem: "hbr-tail-126-genai-copyright"
sourceUrl: "https://hbr.org/2025/07/can-gen-ai-and-copyright-coexist"
sourceTitle: "Can Gen AI and Copyright Coexist?"
---
# Judicial Divergence on AI Fair Use

Early federal decisions in the Northern District of California reveal a stark judicial split on how the **fair use** doctrine (see [[prereq-fair-use-doctrine]]) applies to training generative AI.

In *Bartz v. Anthropic*, [[entity-judge-william-alsup]] ruled that training LLMs on copyrighted books is transformative fair use because the model aims to "turn a hard corner and create something different," likening it to a human reader aspiring to write (see [[quote-alsup-transformative]]). His June 23, 2025 order characterized the use of *lawfully acquired* books to train models as "exceedingly"/"spectacularly" transformative under 17 U.S.C. §107.

Conversely, in *Kadrey v. Meta*, [[entity-judge-vincent-chhabria]] found that unlicensed use is likely *not* fair use, arguing that LLMs are fundamentally different from human learning because they create a product that lets a single user generate "countless competing works" in a fraction of the time (see [[quote-chhabria-competing]]).

**Enrichment caveat (nuance to carry forward):** Legal commentary indicates Chhabria's actual disposition was narrower and partly *procedural* — he rejected the authors' claims because they failed to show reproduction or actual market harm, while stressing that fair use is a holistic, fact-specific inquiry with *market effect* as the most important factor, and refusing to recognize an *automatic* right to licensing fees for AI training absent proven harm. The vivid "countless competing works" phrasing should be treated as a paraphrase of the opinion or secondary reporting unless verified against the Kadrey text. What is well supported is the **existence of a meaningful divergence** in tone and emphasis between an AI-favorable *Bartz* and a market-harm-skeptical *Kadrey*.

This divergence creates massive legal uncertainty and all but guarantees escalation to the Ninth Circuit and likely the U.S. Supreme Court (see [[question-appellate-resolution]]). Relevant doctrinal anchors a downstream expert would invoke: *Authors Guild v. Google* (scanning/search held transformative) and *Andy Warhol Foundation v. Goldsmith* (2023) (new meaning is insufficient if the use occupies the same market function). The piracy dimension is analytically separate — see [[concept-piracy-caveat]].