Nvidia accused of paying for access to pirated content to train its AIs

The world of artificial intelligence is once again shaken by a legal controversy involving Nvidia.

According to recent revelations disclosed in a pending class action, the California-based company did not merely collect data freely accessible on the web, but allegedly sought to pay to obtain privileged access to one of the world’s largest shadow libraries: Anna’s Archive.

The accusations, supported by documents published by the specialized portal TorrentFreak, paint a picture in which the hunger for data to train Large Language Models (LLMs) seems to have outweighed respect for copyright laws.

Nvidia and the alleged deal with Anna’s Archive

At the center of the scandal are some internal communications that appear to pin down Nvidia’s Data Strategy Team. According to documents filed during the discovery phase of the case, company representatives allegedly contacted the administrators of Anna’s Archive to negotiate a “high-speed access”.

Anna’s Archive is known for being a massive aggregator of copyrighted material, including books, scientific articles and texts taken from other pirate portals such as Bibliotik.

What makes the situation particularly critical for Nvidia’s defense is the timing and the awareness of the actions. The emails suggest that, despite being warned about the illegal nature of the collections hosted by the archive, management would have given the “green light” to the operation within a week.

The goal was to download about 500 terabytes of data, an enormous amount of human knowledge necessary to refine the cognitive abilities of generative AIs.

Although there is not yet definitive proof that the payment was actually processed or that the transaction went through, the intent to establish a commercial relationship with a pirate site represents a significant precedent.

A defense that is crumbling

This new wave of evidence has led the authors who filed the lawsuit against Nvidia to modify and significantly expand their complaint.

Initially, the accusation focused on the use of the Books3 dataset, a collection containing thousands of pirated literary works, also used by other industry giants such as Meta and Anthropic.

So far, Silicon Valley’s standard defense has relied on the notion of “Fair Use” (fair use), arguing that training AIs falls into a grey area of copyright law that allows the transformative use of works.

However, the new evidence makes Nvidia’s position much more precarious than its rivals. If it is confirmed that the company actively sought to finance an illicit activity to gain a competitive advantage, the fair use thesis could collapse.

The plaintiffs’ lawyers argue that Nvidia not only used stolen material, but also offered its corporate customers automatic access to contaminated datasets like The Pile, which includes the Books3 collection. This behavior would demonstrate, according to the accusation, total indifference to others’ intellectual property rights.

The irony of intellectual property theft

The case raises a fundamental ethical question that has not escaped industry observers. Companies like Nvidia, which jealously protect their technological patents and trade secrets with armies of lawyers, seem not to hesitate when it comes to appropriating the creative work of writers and researchers.

While Nvidia continues to rake in record profits from selling its graphics accelerators, authors see their works swallowed by machines without receiving any compensation or permission requests.

At the moment, Anna’s Archive remains online, although its growing notoriety has made it the target of ongoing DMCA takedown notices, forcing operators into a continuous cat-and-mouse game to keep the servers running.

The class action against Nvidia, enriched by these new emails, could lead to a turning point in the regulation of AI training, establishing clearer boundaries on what is permissible in the race for digital gold.