Bluesky just made its moderation problem worse
The platform’s own data shows its adult-content classifier has been misfiring at massive scale. The latest update gives it more power.
Editor's note: After this piece was published, @safety.bsky.app announced it is rolling back the account-level moderation changes described here, writing that the changes "missed the mark." The accuracy concerns documented in this piece remain unresolved.
Buried in Bluesky’s May 28 release notes for version 1.122 was a line that’s been generating significant conversation on the platform: the app now supports “expanded moderation labeling at the account level.” For creators who post naturist content or any content that involves non-sexual nudity, the implications go well beyond a UI tweak.
Until now, the primary concern was post-level misclassification: Bluesky’s AI flagging an individual photo as adult content when it wasn’t. That’s been a documented, persistent problem for the naturist community since the platform’s early days. The new expansion means that pattern can now propagate upward, from a post to an entire account.
The scale of the problem is documented in the platform’s own transparency report, published in January. In 2025, the platform applied 16.49 million labels to content, a 200% increase year-over-year. The overwhelming majority, 95.34%, were applied automatically, with no human review. The “adult” label alone was applied 10.79 million times, “suggestive” 3.42 million times, and “nudity” 389,440 times. Bluesky’s own report acknowledges the accuracy problem plainly: with roughly half a million media posts uploaded every day, even a 1% error rate produces thousands of wrongly-labeled posts daily.
This publication deals with these failures regularly. Legitimate news stories, artistic illustration, and stock photography have all been flagged here. Bluesky gives creators the option to self-label content as non-sexual nudity, which is supposed to route it away from the adult content classifier. In practice, the automated system overrides that self-label anyway. Appeals, the supposed human backstop for automation errors, rarely resolve anything. What Bluesky markets as user control is, for a publication covering naturism as a serious beat, closer to security theater. (We’d note that anecdote is ours, not Bluesky’s data.)
None of this is without precedent. The history of how naturist and body-freedom communities ended up on Bluesky is itself a story about automated moderation failures. Tumblr was an early dispersal event. In November 2018, after being suspended from Apple’s App Store over child exploitation material that had slipped through its filters, Tumblr responded by banning all NSFW content, enforced by machine-learning classification. The algorithm proved immediately and spectacularly unreliable, flagging vomiting unicorns, raw chicken, and boot cleaners as adult content within a day of launch. The communities that had built homes there over a decade were erased anyway. Tumblr lost 30% of its monthly traffic within three months and was sold shortly after for less than $3 million, a fraction of the $1.1 billion Yahoo had paid for it. Many of those displaced communities found their way to Twitter, which under its original ownership maintained relatively permissive policies on nudity, though naturists had been present across platforms long before that. When Elon Musk’s takeover made Twitter inhospitable, Bluesky drew a significant wave of arrivals: decentralized, nudity-permitted-with-tagging, and philosophically committed to user control. Naturists built starter packs, established community, and made it work. Meanwhile, Tumblr never solved the underlying problem. As recently as March 2026, its automated system mass-banned dozens of accounts in a single afternoon, with reports suggesting trans women were disproportionately targeted. The platforms change. The pattern doesn’t.
What makes the v1.122 expansion harder to square is that Bluesky’s stated 2026 roadmap explicitly lists improving the accuracy of its automated adult labels as a priority. They’ve acknowledged the system misfires. They’ve committed to fixing it. And then shipped an update that expands its reach.
To be fair, Bluesky’s stated philosophy has always been to label rather than remove, to give users control over what they see rather than make unilateral deletion decisions. That’s a more defensible approach than outright censorship and meaningfully different from how platforms like Instagram or Facebook handle nudity. The transparency report notes that in 96.57% of cases, labels are applied to specific content rather than accounts. The v1.122 update shifts that ratio, if only at the margins.
But for naturist creators, the margins are where the problem lives. An AI system that can’t reliably distinguish a nude body from a sexualized one is now empowered to mark not just a post but a person. The appeals process exists, Bluesky does allow users to contest labels, but at 10.79 million automated adult labels in a single year, volume makes individual appeals functionally inadequate as a systemic remedy.
Bluesky has positioned itself as a principled alternative to the anything-goes chaos of X and the sanitized paternalism of Meta’s properties. That positioning depends on the premise that its moderation is accurate enough to justify its reach. The data from its own report suggests it isn’t there yet. Expanding account-level labeling before fixing that accuracy problem is the wrong order of operations. 🪐
I've tried searching for naturist content on Bluesky and found pornography more often than not.
X isn't 'anything goes" either. I posted a simple nude and it had a "sexual content" label applied.