Progress Interrupted: The Constitutional Crisis in Copyright Law

(The above image was AI-generated by ChatGPT.)

Kevin Frazier is an AI Innovation and Law Fellow at the University of Texas at Austin School of Law.

Imagine a future in which everyone has a good Ph.D.-level research assistant at their disposal. Law professors, medical researchers, philosophers, economists, legislators, educators, and other professionals tasked with generating knowledge could rapidly advance their respective fields. Society would reap the resulting benefits: clearer laws, more effective medical treatments, more personalized education—you get it.

A litany of barriers stands in the way of realizing that future. Technology isn’t one of them. OpenAI’s Deep Research has already proven capable of matching a “good” RA, per leading scholars such as Tyler Cowen. Other research agents will soon follow—if current trends persist, then these agents will have even more advanced capabilities at potentially lower cost.

Copyright laws represent one of those key barriers. AI models require vast troves of high-quality data. Synthetic data—data generated by other AIs—may eventually reduce the need for such data. As of now, good old human-created data is still an essential input to AI training. The median size of training datasets hit 750,000 datapoints in 2023—nearly a seven-fold increase from 2022. Yet, ongoing litigation over AI labs using copyrighted data to train their models has the potential to significantly diminish the quantity of available data. The New York Times (“NYT”), for example, alleges that OpenAI violated copyright law by training its models on the paper’s articles (this case has since been consolidated with copyright claims against OpenAI logged by other publications; a hearing on OpenAI’s motion to dismiss occurred in January 2025). Similar suits have been filed by other publications against other labs.

These cases and related debates, however, miss the fact that the introduction of AI tools like Deep Research mandates a reconsideration of the constitutionality of existing federal copyright laws. This Commentary resurfaces the original purpose of the Constitution’s Intellectual Property Clause and contends that the potential for AI to spread knowledge across the country renders existing statutes unconstitutional by stifling that diffusion.

This bold assertion has significant ramifications on the larger AI governance discourse. The AI race has entered a new phase: DeepSeek’s success in deploying an advanced open-weight model at relatively lower costs than their larger American rivals has leading regulators such as former FTC Chair Lina Khan hyping the importance of fostering open-source model development and competition. The creation of such models depends on many labs having quality training data. Access to such data is by no means guaranteed—OpenAI’s tactics for accumulating model training data have become less technologically feasible as websites work to detect and stop web-scraping bots. For AI development to continue and, more importantly, for the resulting knowledge to spread, copyright law must undergo significant reforms. A full examination of those reforms merits its own treatment. The point of this Commentary is to force policymakers, litigators, and adjudicators to confront how novel AI capabilities demand a reconsideration of the constitutionality of existing copyright law.

The Textual and Historical Meaning of the Intellectual Property Clause

Article I, Section 8, Clause 8 of the Constitution grants Congress the power “[t]o promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries.”

The first clause of this power is not simply hortatory. Professor Larry Solum specifies that the Clause fits into an ends-means structure that courts cannot simply gloss over. Solum’s exhaustive analysis of Congress’s powers under the Constitution makes clear that the Founders intended for Congress to affirmatively seek the end “of promoting the Progress of Science” when crafting copyright law. The Founders were well-versed in distinguishing between preambular statements and clear objectives. If they intended for Congress to merely consider the promotion of the progress of science, Solum observes that they could have and would have penned a clause along these lines: “Whereas the promotion of the progress of science shall advance the general welfare, the Congress shall have the power to . . . “ Importantly, the Supreme Court has reached a similar conclusion. In a 1972 decision, the Court specified that “[t]he direction of Art. I is that Congress shall have the power to promote the progress of science and the useful arts” (emphasis removed). The Court did not merely read over the first clause but considered it as a core part of interpreting Congress’s authority.

Even with that objective in mind, the Constitution does not define what constitutes “progress.” The Supreme Court has yet to offer its own definition. Solum offers the following definition for “progress of science” based on a review of British cases from as far back as the 1500s: “involving advances in learning or the continuation of scientific activity.” In short, the Clause directs Congress to foster the “encouragement of learning.” Professor Malla Pollack leans on public uses of “progress” from the late 1700s and early 1800s, such as stories in the Pennsylvania Gazette, to more precisely define the term. She rejects interpretations of progress that refer to increasing the quantity or quality of science and the useful arts. Instead, her reading of the relevant sources equates “progress” with “spread.” This definition insists that Congress structure copyright law to diffuse and distribute science. Whether existing copyright law achieves that constitutionally-mandated goal has escaped attention in current debates.

Though even more legal scholarship could crystalize the intended meaning of “progress,” the remainder of this Commentary assumes Pollack’s definition has the strongest historical and textual support. While discovery of misplaced notes on the Intellectual Property Clause debate could dislodge Pollack’s contention, given the already impressive amount of scholarly attention to the topic, the adoption of Pollack’s assertion of an equivalency between “progress” and “spread” rests on sturdy ground.

Current Copyright Debates

The primary issue raised in the NYT suit against OpenAI and related cases involving other AI labs provides a useful illustration of the narrow focus of recent suits on AI’s use of copyrighted data. Rather than examine the extent to which copyright law spreads knowledge, litigants have assumed the constitutionality of the existing regime. In the NYT case, both sides have offered contrasting conclusions on whether OpenAI’s use of NYT’s data falls within the fair use exception spelled out in federal law. However, the four-factor analysis associated with that exception evades predictable application. As such, current fair use law makes excellent fodder for legal analysis and debate, popular scrutiny, and seemingly endless litigation.

By focusing on the trees, the legal community has missed the forest. Too little attention has been paid on which individuals and entities truly benefit from copyright law and whether those laws align with the Clause that authorizes Congress to enact such protections. Even when interested parties zoom out to consider broader issues, they stop short of focusing on the Intellectual Property Clause’s clear and unavoidable objective.

For example, consider the Electronic Frontier Foundation’s recent analysis of our current copyright regulatory regime. To their credit, they attempt to return to first principles. They articulate a goal for the broader copyright framework: “encourag[ing] diverse forms of expression from diverse creators who [are] fairly compensated for their role in a profitable industry.” What’s more, they then assess whether that goal is being realized. According to the EFF, it is not. They assert that big media companies, rather than independent creators, benefit most from existing copyright laws. Others have reached a similar conclusion. For example, in their book Who Owns This Sentence?, David Bellos and Alexandre Montagu track the manipulation of copyright law over time to advance corporate interests. Jonathan Tasini, then-president of the National Writers Union, argued way back in 2002 that extensions of copyright protections favor corporations. As noteworthy as this alignment on the effects of copyright law may be, those entities and individuals have failed to ask the foundational question: whether copyright laws help spread knowledge.

Spreading Science (and Knowledge) in the Age of AI

Copyright laws cannot ignore technological advances that alter the means of spreading knowledge. A frozen conception of knowledge diffusion runs afoul of the clear objective set forth by the Intellectual Property Clause. Generative AI tools—even in their nascent stage of development and adoption—have already upended prior channels for spreading knowledge.

AI systems fundamentally transform how knowledge spreads in at least three ways.

First, these systems democratize access to information by breaking down barriers that have traditionally limited knowledge diffusion. Language barriers dissolve when an AI can instantly translate complex research from any language. Geographic limitations become irrelevant when research tools can access and synthesize knowledge repositories from all around the globe. Economic constraints fade when sophisticated research assistance becomes available at minimal cost. This democratization of knowledge access represents precisely the type of “progress” that the Founders envisioned when they crafted the Intellectual Property Clause.

Second, AI accelerates the speed of knowledge dissemination exponentially. What once required months of literature review can now be accomplished in minutes. AI systems can rapidly identify patterns across disparate fields, facilitating cross-disciplinary insights that might otherwise take decades to emerge. This acceleration doesn't merely change knowledge dissemination quantitatively: it transforms it qualitatively. When knowledge spreads faster, more people can build upon it sooner, magnifying innovation across society.

Third, AI enables personalization of knowledge delivery that was previously impossible at scale. Different individuals absorb information differently based on their background knowledge, learning style, and specific needs. AI can tailor information delivery to maximize comprehension and retention, ensuring that knowledge doesn't merely spread but also takes root. This personalization capability means that the same underlying knowledge can be effectively transmitted to experts and novices alike, bridging gaps that have historically limited knowledge diffusion.

These transformative capabilities make AI an unprecedented vehicle for fulfilling the constitutional mandate to “promote the [p]rogress of [s]cience.” According to one leading AI safety researcher, in the unlikely scenario that AI progress slowed, its current rate of advancement will result in AI's collective research capability matching or even exceeding human research capability. Even in that case, technological progress will occur up to ten times faster than a human-only alternative. Put differently, even modest AI progress can help achieve generational technological developments in a span of a couple of years. A more likely scenario is that AI progress will continue to accelerate—bringing about “powerful AI” systems as soon as 2026, per Anthropic’s Dario Amodei. Amodei expects such systems to be "smarter than a Nobel Prize winner across most relevant fields." What's more, powerful AI will "engage in any actions, communications, or remote operations enabled by [a computer] interface, including taking actions on the internet[.]"

Yet current copyright laws threaten to strangle this potential in its cradle. Should the NYT prevail in its litigation, OpenAI might face financial penalties in the billions, while the court could mandate the elimination of all GPT models and training datasets containing NYT content. More generally, by restricting AI systems' access to the data they need for training, existing copyright regimes could effectively prevent the development of the very tools most capable of spreading knowledge throughout society.

To be clear, I am not calling for the elimination of copyright protection. Authors and creators deserve compensation for their work, and incentives for creation remain vital. However, when copyright laws impede rather than promote the spread of knowledge—the constitutional purpose for their existence—those laws have strayed from their legitimate foundation.

Instead, copyright law must undergo thoughtful reform to balance traditional incentives for creation with the unprecedented potential for knowledge diffusion that AI represents. Such reforms might include:

1. Expanded fair use doctrines that explicitly recognize AI training as a transformative use when the resulting systems serve to spread knowledge;
2. Statutory licensing schemes that compensate creators while ensuring AI systems can access necessary training data;
3. Public data commons supported by government funding that ensure all AI developers have access to high-quality training data.

These reforms would not eliminate copyright protection but would realign it with its constitutional purpose in light of technological advancement. The Founders crafted a flexible document precisely because they understood that new technologies would emerge that they could not foresee. They enshrined principles—like the spread of knowledge—rather than specific mechanisms, trusting future generations to apply those principles to new contexts.

It is also worth noting that the Founders did not qualify the spread of knowledge to apply exclusively to human-generated knowledge. Discussions around the Intellectual Property Clause suggest a desire to diffuse knowledge as quickly and broadly as possible, regardless of the source. Tench Coxe, in a 1787 address to the Pennsylvania Society for the Encouragement of Manufactures and the Useful Arts, called for accelerating the adoption of “methods encouraging manufactories” discovered in other countries. This proposal did not consider whether that method involved humans, machines, or a combination of the two. Nor did the proposal discriminate between domestic ideas and foreign ideas. His net goal was to “enable an enlightened society.” Coxe’s remarks likely reached the Founders, given that they were made at the time and place of the Constitutional Convention. A dearth of notes related to the Convention’s debate over the Clause makes his commentary particularly useful in understanding what considerations may have influenced the drafting of the Clause.

Other circumstances increase the odds of the Constitution’s drafters sharing Coxe’s pursuit of the ends, regardless of the means. Prior to the Convention, Madison and others toyed with the idea of offering “premiums”—rewards—to innovators for their inventions. It was even suggested to give up to 1,000 acres for a useful invention. Critically, a land award would have come at little cost to a then land-rich country. Because the immense financial costs of the war left the federal government cash-poor, ideas like land were on the table. That’s also why the Founders were willing to consider time-limited exclusive intellectual property rights despite their fervent opposition to monopolies. The IP scheme outlined in the Constitution required minimal expense to the federal government while still creating some incentive for innovation. The delegates did not hold that scheme out as the best nor even their favored mechanism. It seems that such protections simply amounted to the best the federal government could do at the time.

A contemporary focus on an “enlightened society” cuts against line-drawing exercises about whether humans or AI are responsible for that enlightenment when interpreting the Intellectual Property Clause. That debate, of course, merits broader examination by lawmakers and society as a whole. A strong case can be made for fostering discoveries by human ingenuity primarily. However, that debate is not germane to the application of the Intellectual Property Clause. AI has the potential to drastically accelerate the realization of an enlightened society. So long as that remains the case, the Clause calls for encouraging AI development and diffusion and for removing barriers to that progress.

The Role of the Courts

Some will call on courts to stay out of this debate, arguing that Congress alone can determine what amounts to "the progress of science." This nonjusticiability argument suggests that courts should defer to legislative judgment on how best to fulfill constitutional objectives. However, this position has been undercut in other contexts. For example, the Supreme Court has identified limits on Congress's Spending Clause authority despite similar arguments that Congress alone should determine what constitutes "the general welfare." In South Dakota v. Dole and more recently in NFIB v. Sebelius, the Court affirmed its responsibility to enforce constitutional boundaries on congressional power, even when those boundaries involve subjective determinations.

The same principle applies to the Intellectual Property Clause. While courts should show appropriate deference to congressional expertise, they cannot abdicate their responsibility to ensure that copyright laws actually serve their constitutional purpose. When copyright restrictions actively impede the spread of knowledge—the very thing they are constitutionally mandated to promote—judicial intervention becomes not only appropriate but necessary.

That said, I do not believe courts should be the primary architects of copyright reform in the AI era. Courts are necessarily limited to case-by-case determinations and lack the institutional capacity to design comprehensive regulatory frameworks. Furthermore, many AI-related copyright questions involve complex policy tradeoffs, which are best addressed through legislative deliberation. While courts will likely continue addressing AI issues as lawmakers hesitate (given the rapid pace of technological development and the relatively slower pace of legislation), Congress should take the lead on bringing copyright laws into the AI age. The legislative branch possesses both the democratic legitimacy and institutional tools necessary to enact the massive legal changes required.

Ideally, a balanced approach would emerge: courts would enforce constitutional limits when necessary, while Congress would proactively reform copyright law to both protect creators and harness AI's unprecedented potential for knowledge diffusion. This collaborative institutional approach would best serve the Founders' vision of copyright as a means to an end: the spread of knowledge throughout society.

Conclusion

The constitutional mandate outlined by the Intellectual Property Clause is clear: copyright laws must promote the progress—the spread—of science and useful arts. Today's copyright regime, with its ever-expanding terms and restrictions, increasingly serves corporate interests rather than this fundamental purpose. When copyright law actively impedes the most powerful knowledge-spreading technology in human history, it has drifted from its constitutional moorings.

The legal debates surrounding AI and copyright must expand beyond narrow questions of fair use to confront the fundamental constitutional issue at stake. Courts considering cases like NYT v. OpenAI should evaluate not just whether particular uses fall within current statutory exceptions, but whether current statutes themselves fulfill their constitutional purpose in the Age of AI.

Congress, too, bears responsibility for enacting copyright laws that actually promote the spread of knowledge rather than merely extending corporate monopolies. The historical evidence from scholars like Solum and Pollack provides a roadmap back to the original understanding of the Intellectual Property Clause—one focused on spreading knowledge broadly throughout society.

The Founders could not have imagined AI, but they did imagine a society where knowledge flowed freely to all its citizens. They crafted the Intellectual Property Clause specifically to advance that vision. It falls to us to ensure that our copyright laws serve that constitutional purpose, even—especially—when new technologies transform how knowledge spreads.

A PhD-level research assistant for everyone remains within reach, but realizing this future requires that we reclaim copyright’s constitutional purpose. Only then can we fulfill the Founders' vision of a society where knowledge progresses—spreads—to benefit all.