The DOJ argues that training language models on copyrighted works can be fair use due to transformative benefits, and that limiting such training could stifle innovation and economic growth.

Public benefits from large language model training—advancing creativity, research, and national interests—are asserted to outweigh any potential harm to publishers.

The NYT’s fair-use position is described as potentially restricting innovation and misreading the fair-use doctrine in a way that could hinder AI development.

Publishers and authors are pursuing legal action and settlements as AI training practices come under scrutiny.

Lawsuits seek to decide whether AI training constitutes fair use or requires licensing, with licensing costs potentially shaping industry dynamics.

The New York Times sued OpenAI and Microsoft in 2023, alleging their AI models were trained on millions of Times articles without permission.

The Times lawsuit contends OpenAI and Microsoft trained on millions of NYT articles without authorization.

The DOJ’s analysis focuses on training usage and leaves open questions about data acquisition, storage, and outputs that reproduce protected expressions for future litigation.

The piece weaves in related copyright cases and quotes from the DOJ, aiming to balance AI development with protection of copyrighted works.

Rulings in related cases have centered on fair use and market impact, with the DOJ arguing that outputs must compete to cause market dilution and that training itself is the transformative use.

AI firms broadly claim fair use for training data, while creators across media push back with numerous lawsuits seeking compensation.

Fair-use factors discussed include purpose and character of use, and market effect, with arguments that training does not inherently harm the market and that outputs resembling existing works should not dictate transformativity.