Pornhub promised to ban ‘deepfakes’ videos. And it failed miserably.

In February, Pornhub announced it would no longer tolerate AI-generated pornography featuring the facial features of celebrities over the bodies of adult actresses.

And it’s kept its promise, except… not really.

As Buzzfeed’s Charlie Warzel says:

While banned material frequently slips through the cracks on large sites that allow users to upload content, the deepfake violations on PornHub are especially flagrant. More than 70 deepfake videos were easily searchable from the site’s homepage using the search term “deepfake.” 

What it is: So-called ‘deepfakes’ were born of an algorithm of the same name. It uses training material — mostly from YouTube clips and Google Images results — to arrange pieces on the fly in an attempt to match the source material’s face to that of a video subject. Most commonly, ‘deepfakes’ have been used to create semi-convincing fakes of celebrities appearing in pornography.

Why it matters: While sickening, and undoubtedly harmful for those involved, pornography might not even be the worst of it. Some users figured out early on that the AI could be used for something far more nefarious: propaganda. So far the results have been harmless and laughable (like Buzzfeed’s example below), but it’s not hard to imagine the dangers posed by the likes of Barack Obama, Donald Trump, or Hillary Clinton being “caught” on tape delivering a speech full of words they’ve never said.

Quick take: Pornhub has become the ultimate “you had one job” meme. It succeeded at the most basic level of removing “deepfakes” from its search results, but “deep fakes” and “deepfake” videos still appear. Worse, the company’s algorithms appear to be promoting other non-consensual content with various derivatives of the word “deep fakes.”