Abstract:
The U.S. government recently submitted an opinion to the court in support of artificial intelligence companies using copyrighted content to train large language models without permission. The document believes that if U.S. companies face higher data acquisition costs due to copyright rules, they may fall behind foreign competitors who are not subject to equal constraints in the artificial intelligence competition and have an impact on national security.
This amicus opinion, signed by U.S. Deputy Attorney General Stanley Woodward, was submitted in the context of a copyright lawsuit between OpenAI and the New York Times. The document contends that even if AI training involves copyrighted text, the innovative value and public benefits it brings may outweigh the competitive harm to the market for copyrighted works. The government also stated that if the court makes an overly strict "fair use" ruling, only a few large technology companies will be able to afford the high licensing fees, which will instead weaken competition in the artificial intelligence market.
The relevant position is highly similar to the views previously proposed by OpenAI. OpenAI has argued that if training artificial intelligence must obtain copyright authorization one by one, American companies will have difficulty maintaining the speed of development, and overseas competitors may gain an advantage. The document submitted by the U.S. government this time also believes that artificial intelligence models are changing the economic and national security fields and helping people, including those in the creative industry, complete their jobs.
Critics point out that the current artificial intelligence market itself is mainly controlled by large companies such as OpenAI, Anthropic and Google. Allowing these companies to use other people's works for free may not promote real market competition. At the same time, AI companies have accused smaller startups of using their cutting-edge models without authorization, which may lead to inconsistent applicable standards. Artificial intelligence training does not always produce completely new content, and models can sometimes reproduce original texts or large sections of content in the training data. Some media and creators stated that their works have been directly copied by artificial intelligence systems without authorization and without attribution, which is in clear conflict with the principle of fair use.
New York Times spokesman Graham James said that the government is favoring a few artificial intelligence companies worth trillions of dollars at the expense of a large number of American creators. He believes that artificial intelligence companies and creators could develop together, but the former should pay reasonable fees for the content that supports their products in accordance with the requirements of copyright law. If companies are allowed to use these works without permission and without compensation, it may destroy the ecosystem on which human creative content depends, and the artificial intelligence industry itself relies on this ecosystem to continue to operate.
The U.S. government finally asked the court to reject claims related to "copyright infringement by using copyrighted text to train large language models." However, this opinion is for the court’s reference only and the court may still choose not to adopt it.
The lawsuit began in December 2023, and the case progressed slowly. As of September 2, 2026, the court requires parties interested in submitting amicus curiae opinions to complete their submissions by October 16.
Comments