Adam Mosseri attributes Instagram and Threads' moderation issues, causing users to lose access to their accounts, to human moderator “mistakes”, not AI systems
“Yeah, the humans. AI's gonna save the world!” [embedded post]
That distinction sits uneasily with later reports of users attributing deleted accounts to AI moderation, while Meta cited lower error rates for AI than human reviewers in its response to fresh account-deletion complaints. The recurring issue is less the label on the decision-maker than whether affected users can reliably regain access.
First-order effects
Users who lost access face an attribution dispute: Meta’s public explanation points to moderator mistakes rather than a systemwide AI failure.
Instagram and Threads must treat human-review decisions and appeal handling as the immediate accountability surface, even as they continue using moderation tooling.
Second-order effects
A human-error explanation raises pressure to make enforcement and restoration processes legible to users; otherwise similar incidents will continue to be perceived as automated failures.
Threads’ newer fact-checking rollout adds another enforcement workflow whose handoffs with Instagram and Facebook systems can affect how consistently moderation decisions are applied.
Third-order effects
As platforms blend automated detection with human judgment, moderation quality will increasingly be judged by auditable outcomes—error correction, appeals and account recovery—rather than by whether AI or people made a particular call.
If account-access failures persist across review models, platform trust may shift toward demands for clearer procedural safeguards around high-impact enforcement actions.
The trend: Social platforms are moving toward hybrid moderation, making governance of review, appeals and recovery as important as the accuracy claims of AI systems.
For those of you who've shared concerns about enforcement issues: we're looking into it and have already found mistakes and made changes. Most prominently, our reviewers (people) were making calls without being provided the context on how conversations played out, which was a mi…
I'm glad Mosseri stepped up and owned this, but this is EXACTLY why we have been begging for their definition of political content. There is no way to know if we are playing by the rules when we don't know what they are. So when mistakes happen, we all lose our minds.
I appreciate you coming on and owning this issue. It means a lot. But do you see now why we have constantly been asking for you to define political content so we all know the rules here? Without those, hysteria reigns when we don't hear from you guys.
You're going to lose great content creators here and people who drive engagement. I hope you get this fixed and let people talk about stuff they want to talk about (which includes what's happening in elections)
To clarify, of course reviewers need context, and yes we knew that. One of the tools we built broke, and so it wasn't showing them sufficient context. That's on us.
As Adam noted, the situation with human reviewers was one of the issues and that we continue looking into things. But on the human review point: despite some assertions this was caused by ‘AI run amok,’ it turned out one of our tools was broken so reviewers didn't have the full …
I call bullshit on people reviewers because nearly all of my removed posts were rejected in the same instant as I requested a review. I work fast but I know damn well you don't have enough staff who can read text and pass judgement in half a second after being requested. …
Thank you @mosseri though I have two questions. Why did it take so long to address this issue? If the site was operated by a single person working out of the attic or basement I could possibly understand the delay. Who is responsible for taking care of this?
I was late to post just because it's been a hectic week for other reasons, but that's on me. And, generally speaking, I'm responsible for Threads, so taking care of this is ultimately on me.
“Let's see, who or what should we throw under the bus: humans that we force to deal with this crap or human-created crap software, er, AI?” — “Yeah, the humans. AI's gonna save the world!” [embedded post]