The underlying tech being used by Pornhub could one day be used by more nefarious actors to identify amateur and unwitting porn models.


The robots are watching porn. And amateur porn stars should be concerned.

Pornhub announced that it is using machine learning and facial recognition to detect over 10,000 porn stars across the site in a bid to make searching for your favorite star or fetish more efficient. So far, it's scanned 50,000 videos in the last month of beta testing, and Pornhub plans to scan all five million videos currently hosted on the site by early next year. Eventually—with the help of users, who will upvote and downvote the AI's guesses—the company says it'll get to the point where the computer vision can discern hair color, sex position, and fetish.

Pornhub declined to tell me what computer vision company created the AI, citing a non-disclosure agreement, meaning we don't know if, for instance, the company that created the AI will have access to the data the AI gleans. While the tech seems cool at first glance, and will no doubt be a boon to Redditors shouting "sauce" on every porn gif, there are two major things to consider about this technology: porn piracy and amateur porn star privacy.

Copyright and piracy

Like regular videos, porn clips almost never stay in one place. Just as a single video clip may be posted on YouTube, Vimeo, and other tube sites, porn scenes are often uploaded on Pornhub, YouPorn, xvideos, xHamster, and so on, and they are often uploaded by people who are not the original copyright holders. For instance, the Ted Cruz-approved clip, which was reposted on a porn aggregation Twitter account (and other streaming sites), actually belonged to Reality Kings, whose parent network is Pornhub.

This means that, like the rest of the entertainment industry, the porn industry has a piracy problem. In theory, machine learning could be used to automatically detect when copyrighted content is uploaded and could be used to remove it from Pornhub, creating a system that is similar to YouTube's "ContentID" system, which matches audio and video tracks to identify copyrighted content uploaded by people who don't own it.

For porn actors and companies that produce porn, an AI-fueled content ID system would be a lot more efficient than hunting down every pirated instance of their work and sending a takedown notice to the website owner or host—something independent artists of all kinds deal with constantly, to a distracting and soul-crushing degree.