What do Reddit and many AI companies have in common? Sam Altman’s early involvement and perpetual litigation. In the age of AI, data has become a new currency. Large companies with troves of copyrightable material like Reddit want to protect their valuable data, which AI companies want to use to help train their models. This conflict has given rise to many disputes. Some of these data disputes have ended in a settlement for over $1.5 billion. Reddit has even sued a large AI company, Perplexity, along with several companies accused of data scraping to stop Perplexity from using Reddit’s data to train AI models.
Lawsuits like this have raised the stakes. Companies like Reddit are looking for new legal tools to protect their data. They may have found such a tool from an unexpected source: the Digital Millennium Copyright Act (“DMCA”). The DMCA is a potentially powerful legal tool to combat data scraping by AI companies as I’ll explain. Companies that operate websites hosting large volumes of copyrightable material, such as Reddit, YouTube, Quora, and TikTok, to name a few, are particularly at risk. These companies should take note of how Reddit used this statutory scheme to prevent artificial intelligence companies from improperly using their data. But before we talk about this older law that is being used in new ways, however, some context is necessary.
The Data Problem
AI companies like Perplexity use large amounts of information to “teach” their AI models how to think and respond like a human. The sum of this information—which can be text, audio, images, or many other information media—is called training data. A robust dataset is essential to train a successful AI model since a model without adequate information is of little utility. A major problem for AI companies is how and where they can get large swaths of data to train their models. The internet offers an easily accessible source to obtain training data from compile a complete dataset. Online discussion forums like Reddit provide a veritable goldmine for AI companies.
One problem with taking information from the internet is that the work being downloaded could be copyrighted. A copyright is a legal protection for creative works like a book, song, artwork, or many other forms of authorship. This legal protection is given to the author of the work. For example, the author of an original post on Reddit owns a copyright in the content of their post (subject to a non-exclusive license to Reddit). This protection should prevent other people from copying the work without authorization through the threat of money damages. But, AI companies have been taking content anyway.
With millions of users generating content daily, there is no shortage of training data seemingly up for grabs on the internet. Reddit, however, is not going to let others use their information without a fight.
Copyright Protection Alone is Insufficient
So why does Reddit care about protecting its users’ copyrightable content? Reddit cuts lucrative deals with companies like Google to license its valuable data. Companies want to purchase this data for many reasons including training their own AI models. Here’s the problem: when Perplexity takes data off the internet from Reddit without paying them, Reddit loses out on valuable licensing revenue. Instead of paying, like Google has to, Perplexity gets the data for “free.”
Reddit and those other entities who host large volumes of copyrightable material have been left with limited legal options to fight data scraping by AI companies—when companies scour the internet and take copyrighted material to train their AI as discussed above. Although Perplexity does not have a license to use Reddit’s copyrighted information, copyright protection can be difficult to enforce because of the potential for a fair use defense.
While a whole blog can be (and has been) written about fair use defenses in AI training data, the relevant takeaway for this blog is that fair use is expensive and complicated for a copyright owner to litigate. There is also a lot of uncertainty in rulings from different courts. Some courts have already found that using copyrighted material to train an AI model can be valid fair use. Another court has found training AI models not to be a fair use. The same problem produced two different results, creating uncertainty for copyright owners. The uncertainty and expense of litigating this defense have forced copyright owners to look elsewhere for help. Reddit has found creative solutions to prevent the loss of its data in the DMCA.
The DMCA Solution
Congress passed the DMCA in 1998 to update U.S. copyright law considering the dawn of the internet. The Act gave legal protection to copyright owners who displayed copyrighted material online against unauthorized access to their copyrighted works. One important difference between the DMCA and traditional copyright law is that the DMCA is meant to protect online works and digital media even when traditional copyright rights have not been violated. The DMCA is intended to help ensure that copyright rights are not easily violated due to ease of copying on the internet. Specifically, the statute protects unauthorized access of copyrighted material, not the actual copying of the work. This protection allowed website owners to sue people who circumvented their online protections, such as bypassing password or encryption protection. For example, online copyright owners used this law the late 2000’s to prevent the use of DVD copying software. Reddit is now leaning on the same provision to prevent AI companies from scraping its website’s data.
One important requirement to initiate a DMCA lawsuit is that the website must be guarded by technological protection measures. Basically, a website has to have preventative safeguards to protect copyrighted content in place to sue under the DMCA for unauthorized access.
For instance, Reddit uses a Google product called “SearchGuard” to make sure bots cannot access its data. This is the protection that Perplexity and its codefendant, SerpAPI, allegedly bypassed without authorization. So, Reddit satisfies this protection requirement.
But, you may be wondering how Reddit can sue under the DMCA for unauthorized access of copyrighted material created by other users. Notably, Reddit does not own the copyright of the content that users post. Rather, it obtains from users a broad license. The DMCA does not require ownership of the copyright to protect a host website, only that the website contains copyrightable material. The license itself gives Reddit the right to sue. It is under this principle that Reddit sued Perplexity and SerpAPI for scraping Reddit’s online content through bypassing the SearchGuard product. The beauty of this legal theory is that it does not require litigating the fair use defense.
Such a legal claim certainly has its own challenges, but Reddit has recently found early success in advancing this type of claim under the DMCA. The judge rejected Perplexity’s motion to dismiss the lawsuit and allowed Reddit’s case to move forward. Now, Reddit must back up its allegations by proving what it claimed to be true.
Watch and Learn . . .
While this case is still in its infancy, companies who host copyrighted material online should be proactive about protecting assets such as proprietary data. As this litigation continues to play out, these companies that need to stay tuned to see how arguments on both sides develop. Some of these companies, such as Youtube, have already followed Reddit’s example by filing similar lawsuits. Other hosts of large repositories of copyrighted data should use the DMCA as leverage to monetize their data, or to prevent theft of proprietary data.
In any event, AI companies will continue to amass data on the internet through any means they can. Proactive use of the DMCA’s provisions will provide a thicket of protection for valuable corporate assets. In the digital age, data can be a company’s most valuable property. Going forward the DMCA will be an effective tool to protect data from being misappropriated by other companies.
John Golden
Associate Blogger
Loyola University Chicago School of Law, J.D. 2027