Brian Frye argues AI source credits could benefit newspapers in copyright fight

Giving newspapers credit in AI answers could bring them readers and revenue, Tulane law professor Brian L. Frye argues. But that proposed benefit is not a settled legal answer. In a November 2 interview published by Tulane Law, he outlined a compromise for the OpenAI copyright litigation while acknowledging that courts remain divided over AI training.

Frye said he recently filed an amicus brief—an outside submission offering arguments to a court—in the consolidated New York Times v. OpenAI litigation. He wrote it with journalist and filmmaker Ron Suskind and the CEO of an unnamed small AI company that built a model intended for journalists. The interview explains their position; it does not announce a court decision.

A newspaper’s brand, not just its articles

The brief focuses on market harm, Frye said. Its argument is that training on newspaper articles does not harm their market or value in a legally relevant way because their primary value lies in attribution rather than reproduction. On that view, identifying the sources behind AI output could mitigate harm by directing attention and business back to the newspapers.

That is a claim about how journalism gets its commercial value, not merely a request to label chatbot answers. Frye argues that the court should pursue a compromise benefiting both sides and the public. He also contends that the parties cannot readily acknowledge room for compromise because doing so could weaken their positions in the litigation.

In other words, what newspapers are really selling is their brand, not individual articles.

Brian L. Frye, in the Tulane Law interview

A new purpose does not settle market harm

The underlying dispute starts with copying. As Frye explains it, copyright owners argue that AI companies must copy protected works into training datasets and then use the resulting models to generate competing material. AI companies answer that training serves a transformative purpose: creating a model that produces new works. They invoke fair use, a legal defense that limits copyright protection.

Frye’s own explanation favors that defense. He describes training as analyzing patterns across works rather than using their copyrightable elements. He argues that copyright should serve the public by encouraging the creation and distribution of new works, and that fair use should keep it from blocking useful technologies. Those are the legal and public-interest premises behind his proposed compromise.

But a finding that training has a new purpose does not resolve every issue Frye identifies. He said two district courts have found AI training transformative while also recognizing possible harm to the market for, or value of, the works used. The courts identified different kinds of potential harm for different reasons. His brief therefore concentrates on that second question rather than only defending training’s purpose.

Confidence about the outcome, uncertainty about the path

Frye nevertheless predicts that courts will eventually find AI training non-infringing, probably through fair use. He expects a definitive answer may require the Supreme Court. His reasoning includes the transaction costs of licensing all training data, which he argues could burden U.S. companies while foreign competitors ignore those costs. That is his forecast, not an outcome the courts have delivered.

He also points to the Justice Department’s support for fair use in the OpenAI case. That intervention predates the interview: The Washington Post reported on September 2 that DOJ sided with OpenAI and Microsoft against the Times, arguing that the success of the U.S. AI industry is an important national-security interest. It supports Frye’s account of the government’s position, not his prediction of how the Supreme Court will rule.