Report from MediaPost
In Brief – The US Department of Justice has filed a brief in The New York Times’ copyright lawsuit against OpenAI arguing that training AI models on copyrighted works is generally transformative and fair use under US copyright law because building an interactive AI system differs fundamentally from using articles to inform or entertain readers. The government also argued that training does not harm markets for copyrighted works because the material is not publicly disclosed but instead is used to help models recognize relationships between data and adapt to information. However, the administration did not argue that all OpenAI conduct is protected in all cases, acknowledging that chatbot output may not be transformative when it reconstructs and disseminates copyrighted works.
Context – The proper treatment of AI training under copyright law is being debated globally, but major copyright lawsuits in US courts will be most impactful. The key question is the application of the “fair use” doctrine. As we noted back in March when the Trump Administration released its AI legislative framework, they are firmly behind basic model training being fair use because they believe the alternative would severely hamstring AI development. They directed Congress to leave it to the courts. Two conflicting court opinions released last summer highlight the complexities. Judge Alsup’s vigorous defense of generative AI training as “fair use” was countered the following week by Judge Vincent Chhabria who created the novel concept of “indirect substitution” through which AI systems nullify the fair use defense by creating massive volumes of cheap content that do not actually copy originals but are “similar enough to compete with the originals and thereby indirectly substitute for them”. The DOJ brief criticizes Chhabria’s opinion. If basic training is judged to be fair use in the US, expect other major markets to follow suit in order not to fall drastically behind in development. For example, the UK CMA’s rule requiring Google to give publishers an opt-out for their content not be used by Google’s AI services does not apply to basic model training.