Key Takeaways from AI Progress Panel on Copyright, Fair Use, and America’s AI Advantage

The legal foundation that allows AI to learn from existing works is quietly shaping America’s tech advantage. At a recent AI Progress event on Capitol Hill, experts explained that U.S. copyright law — particularly the fair use doctrine — underpins America’s ability to develop AI models that deliver real societal benefits, from advancing medical research to accelerating scientific discovery and driving the innovations that boost economic growth. 

The big picture:

This enduring legal framework protects intellectual property and drives innovation, enabling access to diverse materials for training AI models. That access is essential. Broad, varied datasets allow AI models to learn patterns that help solve real-world problems. 

• Yes, but: As other nations move quickly to shape the future of AI, changes to U.S. policy that limit data access or disrupt research could weaken America’s competitive position and jeopardize the positive societal benefits AI can deliver. 

Why it matters:

As debates continue over how AI models are trained and what copyright law allows, misunderstandings about AI training are fueling proposals that would restrict the data access needed to build accurate and safe AI systems. 

• “The policies we set today will determine not only the trajectory of AI, but whether the U.S. maintains its leadership in this rapidly evolving field,” Anna Chauvet, a partner at Finnegan and an advisor to AI Progress, said in opening remarks at the Capitol Hill event.

The concern:

Proposals for licensing mandates and other constraints could make AI training prohibitively complex. Licensing material across millions — if not billions — of sources “is a logistical nightmare,” and no centralized system exists to identify ownership or negotiate licenses at scale, Susan Kim, associate general counsel at OpenAI, said at the event. 

• Such requirements would give a competitive advantage to large companies with deep resources while squeezing out startups, researchers, universities, and nonprofits.

The stakes:

U.S. leadership in AI is not guaranteed — and other nations, both allies and competitors, are moving quickly. “If it becomes too risky or too expensive to train models in the United States, development will simply move elsewhere,” Kim said.

Dive deeper:

Beneath recent headlines about lawsuits and licensing deals, a fundamental question is shaping the future of artificial intelligence: how U.S. copyright law applies to the way AI systems learn. At the Capitol Hill event, experts stressed that fair use is a core feature of the copyright framework, enabling AI models to learn from diverse materials in transformative ways that support innovation and U.S. competitiveness. 

1. Training teaches models patterns.

Generative AI models, particularly large language models (LLMs), learn relationships between words, syntax, and concepts across trillions of tokens — small parts of text that represent words, sub-words, or even characters that the model can process. This statistical learning process allows models to generate new responses based on language patterns, rather than retrieving or reproducing specific source material.

“This is the model understanding patterns, not memorizing tiny little copies of all the training data,” said Janel Thamkul, deputy general counsel at Anthropic.

2. Fair use is a built-in feature — not a loophole.

Fair use has been a pillar of U.S. innovation for nearly 185 years, guiding how copyright law applies to new technologies. Courts have repeatedly applied the doctrine to emerging tools that rely on analyzing existing works for new, non-substitutive purposes, from internet search engines to plagiarism detection software.

“Fair use explicitly allows new technologies to learn from, study, and use existing works without replacing them,” Kim said. “Framing it as a loophole just isn't right, because I consider loopholes to be accidental.”

3. U.S. courts are already providing clarity.

Recent rulings in Kadrey v. Meta and Rana v. Anthropic mark an important step in grounding the AI copyright debate in existing legal precedent, with judges examining how AI systems are trained and how copyright law applies to that process.

“The courts found that AI training is a highly transformative process, and after applying the traditional four-factor fair use test, they concluded that AI model training promotes innovation without substituting for the original works,” Chauvet said.

What’s next:

Preserving existing U.S. copyright law — particularly fair use — will ensure America remains the global leader in AI. Maintaining broad access to diverse data is critical for building accurate, safe, and trustworthy models that can drive breakthroughs in health care, science, education, and productivity — all of which contribute to economic growth.

• “Fair use has actually fueled every major digital breakthrough in the last 30 years,” Kim said. “Our country's flexible and balanced copyright system is what creates a safe space for innovation and creativity to thrive.” 

The bottom line:

Policy choices that limit access to training data could slow innovation, narrow competition, and shift AI development away from the United States. Preserving America’s copyright framework will help ensure that AI remains a force for progress — one that expands opportunity, strengthens sectors, and improves American lives.