OpenAI shares court briefs supporting fair use in AI training

OpenAI said it is sharing excerpts from court briefs filed by creators, academics, accessibility advocates and legal experts that argue fair use supports AI training and innovation. The post adds to an active copyright debate in which courts have begun addressing whether training large models can qualify as fair use, while also leaving unresolved questions about data acquisition, retention and market impact.

Evidence and context

The company’s Oct. 30, 2026 post does not present a new court ruling. Instead, it curates arguments from third parties in litigation and policy discussions over how U.S. copyright law applies to AI training. OpenAI’s framing centers on fair use as a legal doctrine that can accommodate training on copyrighted material when the use is transformative and does not substitute for the original works.

Recent court developments show why the issue remains contested. On June 23, 2025, in Bartz v. Anthropic, a court found model training to be fair use on the record before it, but separately rejected fair-use protection for pirated books retained in a central library, according to a Sullivan & Cromwell analysis of the ruling. Two days later, on June 25, 2025, Kadrey v. Meta produced another district-court decision finding AI training to be fair use on the evidence presented, with a Jones Day analysis noting that the plaintiffs had not established market harm. Those rulings are not a nationwide rule and do not determine every future case.

Additional background from the U.S. Copyright Office underscores the limits of any broad conclusion. In a May 9, 2025 pre-publication report on generative AI training, the office said training can be transformative, but model functionality, deployment, unlawful acquisition and market effects remain important. That guidance is not binding law. Ohio State University Libraries also published an update on March 20, 2026 reviewing AI-training rulings and the Anthropic settlement, reflecting continuing legal uncertainty around how fair use applies across different datasets and use cases.

The significance of OpenAI’s post is that it places legal commentary and amicus-style arguments back into the public record while the underlying doctrine remains case-specific. The current evidence supports a narrower point: some courts have accepted fair-use arguments for training on the records before them, but lawful acquisition, storage practices and downstream market effects remain separate questions. No next catalyst date was supplied in the source material.