OpenAI Position on Fair Use and AI Training

Core Argument for Fair Use in AI Training

OpenAI maintains that the training of large language models (LLMs) constitutes fair use under established copyright law. The company argues that training uses copyrighted works for a fundamentally different purpose than the original creations, does not provide the public with a substitute for those works, and does not cause cognizable market harm. OpenAI asserts that copyright is intended to promote progress and creativity rather than to suppress technological innovation or protect existing entities from competition.

Perspectives from Amici Curiae

OpenAI has highlighted several arguments from various organizations and experts who have filed briefs in support of the fair use doctrine in the context of AI development:

• Legal and Academic Perspectives: Law professor Matthew Sag argues that AI training is highly transformative, noting that model outputs are not similar to the intermediate copies created during the training process. Ron Suskind and Brian L. Frye contend that using newspaper articles for training does not harm the commercial expectations of publishers, as those publishers retain the exclusive right to sell access to their original content.
• Industry and Startup Impact: Organizations like Engine and the Developers Alliance emphasize that the purpose of training foundation models is to build general-purpose tools rather than to reproduce training data. The Foundation for American Innovation suggests that the national-security and economic value of new capabilities enabled by AI should be considered when assessing market effects.
• Creator and Accessibility Perspectives: The American Council of the Blind and other library associations argue that restricting fair use would negatively impact individuals with disabilities, educators, and independent researchers who rely on AI tools. Independent creators, such as musician Michael Wall and writer Sarah Rose Siskind, state that AI tools are essential for managing business operations and extending their creative capabilities in a competitive market.
• Technological Progression: Author Kevin Kelly characterizes AI as a new medium that accelerates creativity. He argues that a ruling against fair use would grant copyright holders an overly broad monopoly, hindering the development of technologies that disseminate information and enable new modes of thought.

Market Dilution and Copyright Scope

The Authors Alliance brief addresses the theory of "market dilution," arguing that copyright law has never been intended to protect rightsholders against general competitive pressure. They contend that the law protects against the substitution of protected expression, not against the success of new technologies that merely compete with older works. This sentiment is echoed by Joshua S. Gans, who warns that treating every displaced sale as actionable could weaken the incentive to build new tools without providing meaningful protection to original content creators.

Public Interest and Precedent

Public Knowledge points to historical legal precedents, such as the Google Books and HathiTrust cases, to illustrate how fair use has consistently protected tools that help the public navigate digital information. They argue that applying these established principles to AI ensures that the benefits of the technology remain accessible to the public rather than being controlled by a limited group of stakeholders. The United States government has also weighed in, suggesting that narrowing the fair-use doctrine to exclude AI training would be inconsistent with constitutional principles and would hamper scientific and economic progress.