Post
Very surprised this hasn't blown up yet. But in case you haven't heard, admins of so-far unnamed "popular" federated instances are feeding fedi users' post history to OpenAI for "political ideology analysis" under the delusion that this is an appropriate moderation practice. Surely there are others who might have a problem with this?
@piefedadmin has the details: https://join.piefed.social/2026/05/04/ai-assisted-moderation-in-the-fediverse-is-happening-now-what/
awful systems has defederated from db0 instances for LLM moderation followed by the worst corporate DARVO excuses when caught
@ophiocephalic @piefedadmin it's the db0 group
@ophiocephalic Problem is there's a pretty good chance someone's already feeding AI false data, to further limit its accelerated distribution/oppression.
Need peer reviewed verification.
@numodular
Lots of problems with this scenario. The admins claim it's a "local" Qwen model, but a current GPT model is referenced in the logs. More back-and-forth here: https://piefed.social/c/fediverse/p/2035409/proof-of-ai-assisted-political-profiling-by-unruffled-lemmy-dbzer0-com
@ophiocephalic ouf that's so unethical.
@ophiocephalic
Unfortunate to see this drama pop up on my masto feed. Having read accusations and responses over on Lemmy, this issue looks overhyped. The AI in question is locally hosted. There is no automated scraping. Ironic to me that this mod action seems to disqualify someone from anarchism, while tolerating Zionism does not.
@piefedadmin
@ophiocephalic @piefedadmin the Fascists have been tagging words for a long time now AI will ramp it up. I think I'll start referring to them as assholes Vs. Nazis
@ophiocephalic @piefedadmin Hum, this isn't good. :-|
The fact that the instance admin(s?) responsible for sending the post history of a fedi user to OpenAI for "political ideology analysis" call themselves anarchists is simply tragic. How could anyone with this outlook trust and feed surveillance systems that reify hierarchy and make our worst enemies stronger. This is cop-think. The only conversation anarchists should be having about "AI" is how to most effectively sabotage and destroy this fascist technology
Updated info from @piefedadmin : https://join.piefed.social/2026/05/04/ai-assisted-moderation-in-the-fediverse-is-happening-now-what/
@ophiocephalic @piefedadmin Always assume your online shit is being collected in one way or another.
@yama
I would argue that we don't need to assume or accept that from fediverse admins specifically. We're free to create our own conventions here, and that should definitely include an expectation of the rejection of extractivism from admins
Time for a mass defederation from those instances!
@ophiocephalic @piefedadmin this feels like a defederate-on-sight level breach
@cthon1c
Agree, and @piefedadmin has just updated the post with instance identification
@ophiocephalic @piefedadmin plot twist: anarchist nexus but feeding you into the maw of Leviathan
@cthon1c
a fuckin disgrace
Yeah it is somewhat surprising. @laurenshof this relates to your observation a while ago that many things which would have caused major firestorms a couple of years ago now ... don't.
Although ... the claim is that they're not actually using ChatGPT but instead a locally-hosted model (and the output says ChatGPT as an inside joke), and that they didn't base the decision primarily on the LLM output but rather it corroborated the mod's own judgement. See https://piefed.social/c/fediverse/p/2035409/proof-of-ai-assisted-political-profiling-by-unruffled-lemmy-dbzer0-com and the replies. Messy. One thing's for sure, db0's right in his takeaway that inside jokes don't belong in modlogs.
And this may well explain why it didn't make as much of a splash as it might have: it's Lemmy instances so there isn't as much crossover or awareness.
@jdp23
Couple thoughts. One, zero reason to accept without scrutiny the claim that an actual ChatGPT model is referenced in the logs as a joke. What's funny about that?
Two, your mention of judgement is pertinent, very little of it seems to be in exercise at that instance. Looking at the log, I could see that the modded user in question is a zionist. I was able to make that determination with a quick glance at a handful of the user's posts; the same thing every mod on every instance does when they receive every report.
No statistical fabrication of an "ideology analysis" simulation by an LLM was required, just a few seconds of reasoning by the good old wetware in my own head. What is happening to people that they suddenly believe they are incapable of this and require a dependency on a slop machine to achieve basic thinking?
Whatever the answer, anyone incapable of exercising this kind of judgement should not be a moderator on the fediverse. If a user is trolling and has shit politics block them and move on with life
Install bonfire.cafe
Get the full app experience
