Pirated copies for AI training: Meta uses BitTorrent – but "not seeded"
Meta downloaded entire shadow libraries – via BitTorrent in order to obtain huge amounts of data for AI training. This was not illegal.
(Image: JarTee/Shutterstock.com)
In a legal dispute over whether Meta trained AI models with copyrighted books, the Facebook group has admitted to obtaining the content via BitTorrent, but now insists that it did not publish it itself. Precautions were taken to prevent the downloaded data from being "seeded", according to a statement submitted to the court. At the same time, however, a Meta manager has already testified in court that when downloading the terabytes of data, the settings were configured "so that only the smallest possible amount of seeding occurred". It was therefore not possible to prevent this completely.
Meta defends BitTorrent
The lawsuit brought by Sarah Silverman, Richard Kadrey and Christopher Golden concerns the allegation that Meta used data from LibGen, among others, to train the Llama AI models. This shadow library grants free access to copyright-protected literature; in Germany, access to it is blocked by DNS blocking. According to the plaintiffs, this involves terabytes of data. A meta manager has already admitted to downloading the LibGen catalog via BitTorrent, claiming that alternative routes were too slow. By minimizing or preventing seeding, possible infringements were prevented, the US company now argues.
Videos by heise
It is not yet clear whether Meta will prevail in court with this strategy. The company wants to demonstrate that downloads via torrent are not illegal per se, even when it comes to content from collections such as LibGen. It merely obtained data from a "known online storage facility" that was publicly available via BitTorrent. A quote that has already been cited in court shows that Meta was well aware that the procedure was likely to be problematic. An employee said internally: "Using file-sharing networks on a company laptop doesn't feel right." The attempts to prevent "seeding" also fit in with this.
(mho)
