mouth shut. It followed both Facebook and Twitter in announcing special policies during the chaotic post-election period in 2020 and in suspending then-President Donald Trump after the January 6, 2021, U.S. Capitol insurrection.33 (YouTube contests this description, saying: “We use a variety of ways to convey our efforts to the public, from top executives having tough conversations with the press to blog posts providing overviews on how YouTube works.”) Addressing the ‘rabbit hole’ An exception to YouTube’s avoidance of outside scrutiny has been the “rabbit hole” controversy. Beginning in 2018, journalists and academics reported episodes when YouTube fed users a series of recommendations of increasingly extreme political content. Describing her own YouTube surfing, Zeynep Tufekci, a sociologist who now heads the Craig Newmark Center for Ethics and Security in Journalism at Columbia University, noted that when she clicked on videos of Trump rallies, “YouTube started to recommend and ‘autoplay’ videos for me that featured white supremacist rants, Holocaust denials, and other disturbing content.” After creating a fresh account and seeking out videos of Hillary Clinton and Senator Bernie Sanders, Tufekci received recommendations involving “secret government agencies and allegations that the United States government was behind the attacks of September 11.” YouTube, she concluded, “leads viewers down a rabbit hole of extremism.”34 YouTube terminated the hate-spewing channels frequented by the New Zealand shooter. And even before the slaughter in Christchurch, the company had begun responding to the recommendations predicament. In December 2019, it announced that more than 30 adjustments to its recommendation system had reduced the prevalence of “borderline content,” meaning misinformation, conspiracy theories, and other material that brushes up against, but doesn’t quite violate, the platform’s rules. Over the course of 2019, YouTube said, there was “a 70% average drop in watch time of this content coming from non-subscribed recommendations in the U.S.”37 YouTube hasn't been forthcoming about specific examples of borderline content that it is now suppressing or about whether the same degree of improvement has occurred beyond the U.S., which accounts for only 20% of its user base.38 It also hasn't clarified whether the 70% average drop has proven sturdy. As of May 2020, conspiratorial recommendations had rebounded, diminishing the drop to only 40%, according to researchers at the University of California, Berkeley, who nevertheless called the reduction “an encouraging trend.”39 Without offering any updated statistics, YouTube told us that while “you might see consumption of borderline content go up and down,” the company continues to strive to reduce viewing of such material in the U.S. and globally. Another study, published in 2021 by researchers based at the University of Pennsylvania, found “little evidence that the YouTube recommendation algorithm is driving attention” to far- Subway Shooter: Missed Warning Signs Frank James, the man accused of shooting and wounding 10 people in a New York subway car in April 2022, had posted hundreds of videos on YouTube and Facebook, many of which depicted him making violent threats and other hateful statements. Despite his talk about wanting “to kill people” and “watch people die right in front of my f—king face,” the platforms apparently did not bring James’ videos to the attention of authorities before the mass shooting. His accounts were removed after the attack. YouTube's recommendations are responsible for 70% of all time spent on the platform, and the rabbit-hole phenomenon could have serious consequences.35 The government of New Zealand identified YouTube as a key contributor to the radicalization of a lone terrorist who massacred 51 worshippers in two Christchurch mosques in March 2019.36 HOW YOUTUBE SPREADS HARMFUL CONTENT – AND WHAT CAN BE DONE ABOUT IT 7