Mark Warner, one of the Senate’s most insistent tech critics, turned his attention to YouTube this week after aGuardian reportfound that the company’s recommendation algorithm had surfaced inflammatory anti-Hillary Clinton videos six times more than anti-Donald Trump videos in the lead-up to the 2016 election. “Companies like YouTube have immense power and influence in shaping the media and content that users see,” hewarned in a statement Monday. “I’ve been increasingly concerned that the recommendation engine algorithms behind platforms like YouTube are, at best, intrinsically flawed in optimizing for outrageous, salacious, and often fraudulent content.”

Warner’s concerns about YouTube’s recommendation engine aren’t unfounded, nor are they limited to re-litigating the presidential election. AWall Street Journal investigation this week revealed how YouTube’s algorithms drive confirmation bias by pushing users toward more ideologically extreme content, even when they’re not looking for it. YouTube’s recommendation algorithm, theJournal found, leads users to channels with hyper-partisan perspectives and misleading information. If a user shows a particular political bias, YouTube’s algorithm will recommend videos to that user that echo similar political biases, but with more extreme viewpoints. In a world with an endless supply of ever more conspiratorial content, critics fear it’s a recipe for radicalization.

The issue derives, in part, from YouTube’s ad-based revenue model. Current and former engineers told theJournal that YouTube’s algorithm has been optimized to make the site “sticky”—to keep users coming back with recommended videos they’ll want to keep watching, generating additional ad impressions. And while YouTube’s algorithm isn’t specifically seeking out extremist content to show users, it does look for videos that will bring in traffic and keep users hooked to the screen. Those videos often happen to include sensationalist or conspiratorial content.

YouTube is hardly the only tech company under fire, of course: Facebook has struggled for over a year to tamp down the wave of “fake news” stories that overwhelmed its feed during the last election, and Twitter is looking at ways to lower the temperature on the screaming information wars between rival political factions online. But some questions about YouTube’s algorithm, and its apparent inability to control how it is exploited by bad actors, are unique to the platform. Last year, aseries ofexposés revealed how YouTube’s algorithm had been tricked by a mix of content farms and potential child predators into generating a slurry of violent, creepy, and bizarre video recommendations for young children using the site. Parents might set their child down in front of a kid-appropriate cartoon video, only to find an hour later that the recommendation algorithm had taken them down a rabbit hole of computer-generated, increasingly disturbing iterations. Weeks later, YouTube responded bydeleting some 150,000 videos that appeared to show children in distress.

The problem of extremist and conspiratorial political content is a separate but related issue. In both cases, it seems that YouTube has either lost control of its algorithm—allowing the recommendation engine to trade accuracy for clicks—or is failing to stop content creators from manipulating the system to their own ends.

YouTube acknowledged that such recommendations are problematic. “We recognize that this is our responsibility,” YouTube’s product-management chief for recommendations,Johanna Wright, told theJournal, “and we have more to do.” As of last week, the company is mulling a design change tofight misinformation in the wake of breaking-news events, when YouTube, like Facebook, often sees a flood of conspiracy theories. Such was the case with last month’s fatal Amtrak crash: topresults on YouTube in the hours after the crash included videos from conspiracy theoristAlex Jones, Fox News, conspiracy vloggers, and, further down, a handful of local news affiliates.

YouTube’s overhaul is still very much in its earliest stages, given the difficulty of identifying conspiratorial content without human intervention. In the meantime, YouTube plans to label all videos from state-funded broadcasters, which run the gamut from RT, Russia’s state news organization and a known propaganda outlet, to the Public Broadcasting Service (PBS). But there’s no easy editorial fix for digital companies that manage millions of new posts, videos, and other uploaded content every hour. Facebook was the first tech giant to face national scrutiny for its inability to police content, in the wake of an election cycle that saw the platform hijacked by Russia. But practically every social-media platform that relies on user-generated content faces similar problems when it comes to quality control. While Silicon Valley mastered the ability to keep users engaged, far less thought ever went into the consequences. YouTube’s own struggle to diagnose, let alone treat, the underlying illness is a worrisome sign that the brightest minds in tech are still playing catch-up when it comes to their creations.