Setting the Record Straight About America’s Copyright and AI Policy
With U.S. policymakers and judges considering the very important question of how copyright law applies in the context of AI, especially AI training, there’s been an unfortunate deluge of falsehoods and myths spread by AI developers and others in a poor attempt to get policymakers and courts to grant them special copyright treatment for AI. Below, we debunk some of these myths with the actual facts to illustrate why strong copyright protections are essential if the United States wants to dominate in AI.
FACT: Strong copyright protection isn’t the enemy of AI Innovation, it’s the backbone.
Many AI companies claim that the only way they can dominate in AI is for policymakers to grant them special treatment under copyright law so that they can get free and unbridled access to copyrighted works to train their AI systems. Nothing could be further from the truth. For American AI companies to dominate over their counterparts in other nations, America must continue its long history of strong and effective protection for American intellectual property to support the human creation of copyrighted works and not create new exceptions that allow AI companies to use these works without permission or payment.
The United States has a long history of leading the world in innovation and creativity. This leadership is the result of, and is dependent on, established and strong U.S. intellectual property laws. The global protection of U.S. intellectual property is an integral part of ensuring U.S. economic competitiveness, national security, and sustained global leadership. It’s a principle that the first Trump Administration championed in 2020 under its Artificial Intelligence for the American People plan, which listed respect for intellectual property as a core value.
Just take a look at the incredible impact of copyright law on the U.S. economy and job creation. According to the latest report by the International Intellectual Property Alliance (IIPA), the core copyright industries added over $2 trillion of value to the U.S. GDP, accounting for almost 8% of the entire U.S. economy. Sales of major U.S. copyright products in overseas markets amounted to over $272 billion and employment in the total copyright industries was over 21 million workers across all fifty states, representing nearly 10% of the total U.S. workforce. It’s clear that copyright industries contribute massively to the U.S. economy and job creation. Those contributions must not be jeopardized by forcing copyright owners to subsidize AI development.
Over the decades, America’s core principle of strong creators’ rights has produced the world’s strongest and most dominant creative economy and the world’s most innovative and effective high-tech economy. Promoting strong copyright policies ensures the continued success and world dominance of both of copyright and AI U.S. industries.
It is well-known that inputting high-quality copyrighted works results in better AI output, and so continual production of such works is necessary to continue AI dominance in the future. Without human-created, copyright-protected works, AI companies would not be able to develop the models that sustain America’s AI industry. Those models cannot be further developed, and American’s AI dominance cannot be sustained, if U.S. copyright laws are disregarded or eliminated.
There is no reason for U.S. policymakers to pick a “winner” or “loser.” Attempts to do so will create a lose/lose situation that undermines incentives to create new creative works—jeopardizing both the culture and the long-term development of future AI models looking for fresh works to use. Both the AI and copyright industries can thrive if the free markets that they empower continue to be respected.
FACT: America’s advantage is and always has been our intellectual property system. It must not be cast aside to “keep up” with China. We want to beat China, not be China.
AI companies argue that in order to keep up with and beat China in AI, the United States needs to disregard or change U.S. copyright law. But doing so would mean abandoning the principles that have set us apart from undemocratic countries that have no respect for intellectual property rights and have made U.S. creative industries and the works they produce the envy of the world. America’s advantage is and always has been our intellectual property system. It must not, and need not, be cast aside to “keep up” with China.
Casting aside copyright to ensure the success of the U.S. AI industry would inevitably have the opposite effect. Copyright is a primary tool in preventing bad actors from other countries from stealing our nation’s intellectual property. If the U.S. sets the global standard by requiring respect and adherence to copyright laws through free-market licensing, that gives U.S. AI companies an inherent advantage over countries, like China, that do not respect copyright, as AI companies from these countries will produce inferior products that will be unwelcome and illegal in countries that respect copyright.
American businesses thrive in an environment of fair competition; where everyone plays by the same rules. “Deleting IP” or any other shortcut that lets rivals take valuable American intellectual property for free would be a gift to China—which leads the world in mass stealing of creative works. The only way to win the race for AI innovation is not to emulate lawless mass copying and infringement and, instead, stick to our core principles and enforce strong intellectual property against China and every country that doesn’t respect creators’ rights.
On a level playing field, no other country can match America’s second-to-none workers, artists, and innovators. America will dominate AI by leveraging these advantages, not wastefully and unilaterally giving them up. America’s creative economy and cultural works are the envy of the world and provide a powerful strategic advantage to U.S. developers who know how to effectively license them (and in many cases already do so on their existing platforms). Handing this valuable strategic resource reserve to the rest of the world for free under the guise of “fair use” or some other legal fig leaf would undermine U.S. AI leadership, not promote it. No other nation gives rivals its valuable resources for free, why would we?
FACT: Promoting our strong, time-tested copyright laws is integral to the national security interests surrounding the development of AI.
AI companies argue that limiting the use of copyrighted material for AI training amounts to a national security threat because it could impede AI innovation in the United States, potentially granting geopolitical competitors like China a technological advantage. Promoting strong copyright laws and the Constitutional guarantees of the protection of human creators is critical to maintaining American AI dominance, which in turn will diminish national security threats. Ronald Reagan, and many of our country’s great leaders before and after him, have taken a “peace through strength” approach to national security. That tenet is no less applicable to the strength of our intellectual property system. Promoting our strong, time-tested copyright laws and the human creators, copyright owners, and the creative industries they protect is integral to the national security interests surrounding the development of AI.
Global dominance of American culture and arts is a key tool for spreading American values and advancing democracy and freedom. The global popularity of American books, movies, music and art provides vehicles by which we are able to project American values to the rest of the world. Tearing down American creativity leaves us and the rest of the world vulnerable to Chinese propaganda and is an obvious step in the wrong direction.
Before suggesting an unwarranted upheaval to U.S. copyright law for the purposes of national security, AI companies should clean their own house. For example, it has been widely reported that China has been covertly using “ChatGPT to spread propaganda, manipulate social media engagement, and target journalists and politicians in a coordinated AI-powered influence campaign.” And ChatGPT is not alone. A report issued at the end of June by the American Security Project concluded that:
“The Chinese Communist Party’s aggressive censorship laws and disinformation campaigns have resulted in a proliferation of propaganda and censorship across the global AI data marketplace. AI-powered chatbots in the United States now regurgitate CCP propaganda in Chinese and English when prompted on certain topics, posing significant ramifications for global AI development and U.S. national security.”
If AI companies want to talk about national security, how about we start there—a real threat to our national security, instead of the contrived threat of copyright law.
FACT: There is a simple solution that would result in a win-win for AI companies and copyright owners—it’s free market licensing. Licensing is not impossible. In fact, there is already a robust licensing market.
AI companies argue that it is impossible for them to license the copyrighted works that they need to train their AI models and a requirement to do so would halt progress on AI and lead to a parade of other horribles listed above. That argument defies logic and the simple facts.
In the recent case of Kadrey v. Meta, Meta said that a ruling that required them to license the copyrighted works they use for training would “stop this technology in its tracks.” The judge in the case called Meta’s argument “ridiculous.” He explained that AI products are expected to generate $460 billion to $1.4 trillion over the next ten years for Meta alone and thus, “[i]f using copyrighted works to train the models is as necessary as the companies say, they will figure out a way to compensate copyright holders for it.” He added that the suggestion that AI development would halt “doesn’t pass the straight face test.” Most others agree.
AI companies pay for energy, workers, software licenses, servers, and all the other inputs their business and technology require. Obviously, they must also pay market value for the creative works they use to develop commercial products. And the facts show that licensing is occurring in many cases. There is already a robust, emerging market for copyrighted works to be used for AI training. We are already seeing numerous AI training deals between rights holders and developers and most of the big AI developers have already licensed many creative works for their existing products and platforms.
What America does best is innovate, so let the market innovate solutions to these licensing issues. These market innovations are already happening. Numerous startup companies, like Created by Humans, Protégé and ProRata.ai (typically called collective rights organizations), have developed in the past few years to address licensing needs and broker deals between copyright owners and AI companies. If copyright-protected works are essential to AI companies and their businesses, they should be treated no differently than other business essentials that AI companies pay for, like server space or computing technologies.
Securing licenses reduces infringement risks and liabilities for companies. Particularly for AI startup companies, reducing risk and liabilities through responsible practices is a valuable asset can actually attract investors. Free market licensing is far from impossible. It’s the solution, one that should be embraced by AI companies and policymakers across the world.
FACT: The fair use doctrine is applied on a case-by-case basis. There is no such thing as a use that is categorically fair use. Typically, the use of copyrighted works for commercial AI training purposes that harm existing or potential markets for original works is unlikely to be considered a fair use.
Many gen AI companies claim AI training is always “fair use”—but fair use is a fact-specific, case-by-case determination. It allows the use of copyrighted materials without a license in limited circumstances, such as certain types of criticisms, commentary, news reporting, teaching, scholarship, and research. As a recent widely praised, exhaustive Copyright Office analysis makes clear, any blanket claim that “all training is fair use” is plainly contrary to law.
While fair use is decided on a case-by-case basis, there are certain factors that should weigh heavily against a holding of fair use in any case of AI training. For instance, GAI systems are especially unlikely to be qualify as a fair use when they:
- generate outputs that serve the same purpose as, or serve as a direct in indirect substitute for the original creative works they are trained on;
- use pirated material from illicit sources to train;
- are commercial ventures; or
- harm an existing or potential markets for training the original creative works.
One or more of these factors can be found in many of the over fifty cases filed in the federal courts.
FACT: AI models should be developed in a manner that is responsible and ethical. Allowing AI companies to scrape pirate websites, circumvent paywalls, and copy pirated works from shadow libraries is irresponsible and unethical and must not be permitted as a fair use.
Many AI companies scrape pirate sites, go behind paywalls, and even copy pirated works from so-called “Shadow Libraries,” like Books3 and Z-Library. AI models should be developed in a manner that is responsible and ethical. Responsible and ethical development and deployment of AI models means, in most cases, licensing creative works for training from creators—not plundering stolen libraries from criminal websites hiding in overseas jurisdictions beyond the reach of U.S. law.
In any other industry we wouldn’t tolerate illegal raw material undercutting U.S. manufacturers and we shouldn’t tolerate it in the context of AI training either. For instance, when China flooded the U.S. market with artificially cheap steel, the answer wasn’t to give away American steel. Instead, the U.S. government responded with a combination of tariffs and investigations into unfair trade practices. American creative genius should be treated with at least the same level of respect as metal ores.
Copyright protections do not disappear for a creative work just because the work is made accessible online. The availability of a copyright-protected work online does not mean that the work can be used or scraped, unless permitted by the copyright owner or by law. If a book is put in a bookstore display, a shoplifter is not allowed to simply take the book without the proper payment just because the book was made accessible and available.
FACT: AI companies are not just merely copying unprotected data; they are copying copyrighted creative expression.
AI companies claim that they are only using the “functional elements” or “non-expressive elements” of copyrighted works and not using the copyrightable expression. No court has accepted this absurd argument. In the recent case of Bartz v. Anthropic, the judge noted that “[c]opies selected for inclusion in training sets were selected because they were complete and because they contained rich protectible expression.” Similarly, the judge in the aforementioned Kadrey case rejected Meta’s argument that it “only used the plaintiffs’ books to gain access to their ‘functional elements,’ not to capitalize on their creative expression,” explaining that “Meta’s use of the plaintiffs’ books does depend on the books’ creative expression.” As Meta itself notes, LLMs are trained through processing “‘statistical relationships between words and concepts’ and collecting ‘statistical data regarding word order, frequencies [what words are used and how often], grammar, and syntax.’ Word order, word choice, grammar, and syntax are how people express their ideas.”
It is that type of expression that makes literary works (like the ones at issue in Bartz and Kadrey) protectable under copyright. The same is true for music, images, audiovisual works, and other copyrighted works. AI systems typically copy entire works—the whole thing, including all the expressive elements of a work that merit copyright protection. The expressive elements are the value. The argument that AI companies may extract and retain data about those works, while later discarding the copies of the works themselves, does not change the fact that they copied entire copyrighted works—an act that typically constitutes copyright infringement, absent a valid defense.
If you aren’t already a member of the Copyright Alliance, you can join today by completing our Individual Creator Members membership form! Members gain access to monthly newsletters, educational webinars, and so much more — all for free!