Instagram and Facebook users report that Meta's automated moderation wrongly banned real accounts built over years — including at least one business account with nearly a million followers. Worse: the appeal against the ban is also typically decided by AI, leaving users with no path to reach a human. Some cases were only reversed after journalists reached out to Meta directly.
What Changed
According to reporting from TheNextWeb, Startup Fortune, and NotebookCheck, more than 60,000 users have signed a petition asking Meta to explain the bans and allow human review of appeals. Meta, for its part, says its newest moderation tools make 13% fewer mistakes and catch 10% more real violations than human reviewers — and says the cases examined came from older moderation systems, not its current AI.
Why It Matters
The central point isn't whether Meta's AI is more or less accurate than a human — it's that when the system gets it wrong, there's no real human escalation path. A low error rate still produces a large absolute number of wrong decisions at scale, and without human review available, every error becomes an unsolvable problem for the person affected.
The Impact for Brazil
Companies automating decisions that affect customers, partners, or users — credit approval, content moderation, account suspension, support triage — carry the same structural risk Meta is now facing publicly: no matter how good the model's average accuracy is, without a genuine human appeal path, a single error turns into a reputational crisis and, in some cases, a legal problem.
Entercast's Take
Model accuracy is a lab metric; governance is what decides whether a company survives the error that, statistically, will always happen eventually. No decision-automation project should go to production without a genuine human escalation path for the cases where the AI gets it wrong — not as a rare exception, but as part of the process design from day one.