Fair Use & AI
Fair use and AI refers to the application of the fair use doctrine (17 U.S.C. § 107) to AI training and output. AI companies argue that scraping copyrighted works to train AI models constitutes fair use — a transformative, non-expressive use that does not harm the market for the original. Copyright holders argue it is commercial infringement that directly substitutes for licensing. The four-factor fair use analysis applies, but courts have not yet issued definitive appellate rulings on AI training specifically. The Supreme Court's 2023 narrowing of transformative use in Warhol v. Goldsmith significantly weakened the AI industry's primary fair use argument.
The outcome of AI fair use litigation will determine whether the current AI industry model is legal. If training requires licenses, every major AI company faces massive retroactive liability and the economics of the AI industry change fundamentally. If training is fair use, rights holders permanently lose control over how their works are used to build competing AI systems without compensation. No issue in current entertainment law has higher financial stakes — or more direct implications for how entertainment attorneys advise clients on licensing, clearance, and IP portfolio strategy.
The fair use statute — courts weigh (1) purpose and character of use (including whether transformative), (2) nature of the copyrighted work, (3) amount and substantiality of the portion used, and (4) effect on the market for the original. All four factors must be weighed; no single factor is determinative.
The Copyright Office rejected a blanket fair use exemption for AI training, requiring case-by-case analysis. Confirmed that the commercial nature of AI training weighs against fair use under the first factor.
Whether AI image generator training on copyrighted artworks without license constitutes fair use.
Fair use was not resolved at the pleading stage — the court found plaintiffs adequately alleged infringement, preserving the fair use question for trial. Most significant pending AI fair use case.
Whether OpenAI's training of GPT models on literary works without license constitutes copyright infringement, and whether fair use applies to large-scale commercial training on literary content.
Case in discovery — will be dispositive for how fair use applies to LLM training on literary works. Outcome directly affects every AI company using text-based training data.
Whether Warhol's commercial licensing of silkscreens derived from Goldsmith's photograph constituted fair use under the transformative use doctrine.
The Supreme Court significantly narrowed transformative use — holding that commercial licensing of a derivative work serving the same market function as the original is not fair use. Widely cited in AI training cases to weaken the transformative use argument for commercial training.
Whether Google's scanning of millions of books for its Books Search database constituted fair use.
The Second Circuit held that scanning for search indexing was transformative fair use — AI companies argue this precedent supports training data use. Rights holders argue training is commercial and substitutive in a way that book scanning was not.
The AI fair use question has bifurcated the entertainment industry into two camps that often exist within the same company: the IP rights holder arm (that wants training to require licenses) and the technology deployment arm (that wants training to be fair use). Studios, labels, and publishers are simultaneously litigating against AI companies and signing licensing deals with them — a pragmatic hedge against litigation uncertainty. Independent creators are almost entirely in the litigation camp but have the weakest individual legal position. The Warhol decision has meaningfully shifted the landscape against AI companies' primary defense, making licensing deals more attractive as a risk management strategy.
Advise rights holder clients that the Warhol decision is their most powerful current argument against AI training fair use — emphasize the commercial, market-substitutive nature of training in any licensing negotiation or litigation strategy.
For AI company clients, the post-Warhol landscape makes blanket fair use reliance significantly riskier — recommend proactive licensing as a compliance strategy.
When evaluating whether a specific AI tool creates infringement exposure, the fourth factor (market effect) is often determinative — assess whether the AI output competes with or substitutes for the training data source.
Document any licensing revenue your rights holder clients have received for AI training use — this establishes market value and strengthens the market harm argument in litigation.
Monitor Authors Guild v. OpenAI closely — it will be the first major case to directly address LLM training on literary works, and will reshape fair use analysis for text-based AI training.
No appellate court has yet ruled definitively on whether AI training on copyrighted works constitutes fair use — the law is genuinely unsettled.
The Andy Warhol decision (2023) significantly narrowed transformative use, making it harder for AI companies to argue commercial training is transformative.
All four fair use factors must be weighed — AI companies' arguments are strongest on factor one (non-expressive use) and weakest on factor four (market effect on licensing markets).
Licensing is emerging as the de facto compliance standard — major AI companies are signing content licensing deals even without definitive infringement rulings against them.
The outcome of Andersen and Authors Guild v. OpenAI will define entertainment law practice and AI industry economics for years.
Legally unresolved. AI companies argue training is transformative because the model learns patterns rather than reproducing the original. Rights holders argue it is commercial use with direct market effects on licensing. No appellate court has ruled on AI training specifically, and the Supreme Court's Warhol decision weakened the transformative use argument.
AI companies argue yes — both involve processing large volumes of copyrighted works for non-expressive purposes. Rights holders argue no — Google Books allowed searching and snippets without enabling creation of competing works, while AI training enables direct commercial competition. Courts will need to address whether the distinction matters.
Register copyrights, document any AI training use of their content, evaluate licensing opportunities to establish market value, and monitor class action litigation for potential participation. Joining a class action, if available, may be the most practical individual enforcement mechanism.