What Trump’s AI Action Plan Means For Copyright

President Trump’s recently-unveiled AI Action Plan conceptually attempts to address the tension between the rights of copyright owners to control their works and the need of AI companies to use copyrighted works to train their systems. Other solutions, some pro-copyright owner and some not, have arisen to try and address the problem.

Federal Courts find Fair Use

Two federal courts have ruled that the use of copyrighted books to train AI is fair use.

Getty Images

Whether AI companies must secure permission from copyright owners to use their copyrighted works to train generative AI models continues to be a murky and debatable issue. In two recent federal court rulings on the issue, federal judges in the Northern District of California ruled that the use of copyrighted books to train AI systems – Anthropic’s Claude system and Meta’s Llama system, respectively – was a fair use and therefore did not require the book authors’ permission. Those decisions, however, are not controlling outside of their jurisdictions, and, more importantly, are on or subject to appeal. Therefore, they could be reversed – although in my opinion, they will not be. Thus, they do not provide any definitive answer.

Moreover, those decisions, like all court decisions, are limited to their facts. Other AI models, which use copyrighted works differently than Claude or Llama, might require different legal outcomes. Of note, Universal Studios and Disney are currently suing Midjourney for using their copyrighted works, alleging facts that seem much more troublesome than those involved in the Anthropic and Meta suits.

President Trump’s Proposed Solution

The Trump administration favors the fair use position. President Trump has just released an AI Action Plan that prioritizes building the country’s AI capabilities and removing regulatory and other barriers to that end. Speaking at a recent AI Summit, the President said: “You can’t be expected to have a successful AI program when every single article, book or whatever you’ve studied you’re expected to pay for. We appreciate that, but you just can’t do that because it’s not do-able. And if you’re going to try and do that, you’re not going to have a successful program.” Echoing the analysis of Judge Alsup in his fair use decision, which analogized reading a book to increase one’s knowledge to using a book to train an AI system, the President said: “When a person reads a book or an article, you’ve gained great knowledge. That does not mean that you’re violated copyright laws or have to make deals with every content provider,” he said. “You just can’t do it. China’s not doing it.” How exactly the administration will implement such a rule, whether it will, and what authority the AI Action Plan would have remains to be seen.

Legislative Solutions

Meanwhile, on July 21, Sen. Josh Hawley (R-MO) and Sen. Richard Blumenthal (D-CT) introduced a bill that would require AI companies to secure permission from copyright owners before using their works to train AI systems. The AI Accountability and Personal Data Protection Act would create a private tort action against any company using copyrighted material to train an AI system without the copyright owner’s permission. The bill also contains provisions that any agreement to the contrary, other than a collective bargaining agreement, would be void.

Market Based Solutions

Some AI companies are striking deals to compensate copyright owners – or at least the companies who control copyrighted works – for using their works to train AI systems. Examples include a deal struck between Amazon and the New York Times, and between Open AI and News Corp. and the Associated Press.

Opt-Out Solutions

Other AI models have instituted “opt-out” features in their end user agreements or user settings, allowing users to opt out of allowing the model to use its own creations to further train itself. Indeed, laws in countries outside the U.S., such as in the EU, have laws that expressly allow rightsholders to reserve their rights in their work from data-mining, effectively an opt-out of AI data training. Article 4(3) of the 2019 Directive on Copyright and Related Rights in the Digital Single Market states: “The exception or limitation provided for [purposes of text and data mining] shall apply on condition that the use of works and other subject matter referred to in that paragraph has not been expressly reserved by their rightsholders in an appropriate manner, such as machine-readable means in the case of content made publicly available online.” Given this, I noted with interest that DreamWorks used the following disclaimer in the credits of its recent film The Bad Guys 2: “ALL RIGHTS IN THIS WORK ARE RESERVED FOR PURPOSES OF LAWS IN ALL JURISDICTIONS PERTAINING TO DATA MINING OR AI TRAINING, INCLUDING BUT NOT LIMITED TO ARTICLE 4(3) OF DIRECTIVE (EU) 2019/790. THIS WORK MAY NOT BE USED TO TRAIN AI.” Whether this opt-out is or will be legally effective under U.S. law remains to be seen.

The copyright/AI wars continue. Stay tuned.