In a statements Friday, Twitter CEO Jack Dorsey and Twitter’s Safety team revised its rationale for allowing three anti-Muslim videos retweeted by U.S. President Donald Trump on Wednesday. After initially defending the videos as “newsworthy” despite their apparently hateful intent, Twitter now says the videos are permissible under its current policies.
We mistakenly pointed to the wrong reason we didn’t take action on the videos from earlier this week. We’re still looking critically at all of our current policies, and appreciate all the feedback. See our safety calendar for our plans and ship dates. https://t.co/yGytH3eskM
— jack (@jack) December 1, 2017
https://twitter.com/TwitterSafety/status/936669072766332928
Twitter’s current policy classifies as “hate speech” only content whose “primary purpose” is promoting violence against racial, ethnic, or other identity groups.
But Twitter management could just as easily have decided the tweets meet that standard. The videos were shared by Trump on Wednesday, when he retweeted them from the account of Jayda Fransen. According to Slate, Fransen is part of a British far-right movement, Britain First, linked to the assailant in the 2016 assassination of pro-European MP Jo Cox. That association was dire enough to trigger condemnation from conservative U.K. Prime Minister Theresa May, who described Britain First as a “hateful organization.”
The backtracking would seem to put Twitter on better footing going forward in one sense. Arguing that the President’s tweets can’t be filtered because they’re “newsworthy” would be as self-fulfilling as it is self-defeating: everything the President tweets is newsworthy, potentially giving him carte blanche to use the platform however he sees fit.
Get Data Sheet, Fortune’s technology newsletter.
Twitter’s decision is troubling, though, in its implicit permissiveness towards hate. The posts Trump retweeted do not explicitly advocate anti-Muslim violence. But they are part of a broad campaign to malign both Muslims broadly, and recent European migrants in particular, through images or videos presented with little or no context aside from taglines that describe them as crimes by Muslims.
Those taglines are often both inflammatory and factually incorrect. For instance, the Dutch Embassy helpfully took the time to explain the reality behind one of the videos President Trump retweeted.
.@realDonaldTrump Facts do matter. The perpetrator of the violent act in this video was born and raised in the Netherlands. He received and completed his sentence under Dutch law.
— Netherlands Embassy in the US 🇳🇱🇺🇸 (@NLintheUSA) November 29, 2017
Twitter’s re-read of its own policies suggests just how quickly social media outlets, which long resisted responsibility for content shared by users, have had to adapt to rising public pressure to police hate speech, violence, and harassment. That pressure has mounted greatly since Trump’s own election, which some argue was fueled by inflammatory or misleading content similar to the recently-shared video posts.
But, while companies like Facebook and YouTube are taking action, identifying and controlling such content is a major challenge. Companies are, clearly, still struggling to set consistent internal standards. And the huge scale of the platforms suggests that enforcing the rules will require some form of algorithmic or AI-based monitoring.
But the latest Trump flap shows just how difficult that will be. The videos in question are mostly only moderately violent, and the implicit condemnation of Muslims hinges on context – including Fransen’s political affiliations – that might be difficult for a machine to parse. That means a safer and saner social media landscape may rely, at least in the medium term, on major new spending by companies like Twitter – which already has its share of business challenges.
