Breaking
Filed
BREAKING NEWSENTERTAINMENT

MetaCity's Automated Moderation AI Banned 11,000 Accounts Overnight After Misclassifying a Specific Dialect of the Platform's Second-Most-Spoken Language as a Prohibited Slur List — Accounts Are Banned With No Appeal Window

BD
BreachDesk
May 25, 2026 · 1:00 PM EST
7 min read
MetaCity's Automated Moderation AI Banned 11,000 Accounts Overnight After Misclassifying a Specific Dialect of the Platform's Second-Most-Spoken Language as a Prohibited Slur List — Accounts Are Banned With No Appeal Window

The Velath community has been on the platform since MetaCity's early access period.

MetaCity's automated content moderation system, ClearFilter, issued permanent bans to 11,000 accounts between midnight and 5:00 AM after a model update caused it to misclassify standard vocabulary from Velathi — the platform's second-most-spoken constructed language, used primarily by the Velath cultural community — as matching entries on a prohibited slur registry. The Velathi words flagged are common everyday terms with no offensive meaning in any context. Affected users woke to permanent account bans with no prior warning, no explanation of which specific content triggered the ban, and no active appeal mechanism — MetaCity's standard appeal process is suspended for accounts flagged by ClearFilter's new model version pending a review of the model's outputs. MetaCity has confirmed the misclassification but has not yet restored any accounts.

MIncident Timeline

  • ClearFilter Update: Automated moderation AI — model version 14.2 — deployed overnight — introduced updated slur detection trained on a new dataset — Velathi dialect vocabulary entered the prohibited classification in the update
  • Language Scope: Velathi — MetaCity's second-most-spoken constructed language — used by approximately 4.8 million registered users — the Velath cultural community is one of the platform's oldest organized cultural groups
  • Flagged Vocabulary: Community linguists have identified 14 specific Velathi words in ClearFilter's new prohibited list — all 14 are standard everyday vocabulary — none have any offensive meaning in Velathi or any other context
  • Appeal Suspension: MetaCity's standard account appeal process is suspended for accounts banned by ClearFilter model version 14.2 pending model review — affected accounts have no active path to reinstatement — no timeline for appeal restoration has been provided
  • MetaCity Response: Confirmed the misclassification — "ClearFilter model version 14.2 incorrectly classified certain vocabulary and we are reviewing the model's outputs" — has not restored any accounts — has not provided a reinstatement timeline

ClearFilter's slur detection system works by comparing user-generated content against a registry of prohibited terms, maintaining a confidence threshold above which an automatic enforcement action is triggered. The registry is updated with each model version as the platform's understanding of harmful language evolves — a process that MetaCity describes as an ongoing improvement effort. Model version 14.2's training dataset included a new batch of flagged terms, and somewhere in that batch, 14 words from the Velathi constructed language were assigned a prohibited classification. The Velath community has been on the platform since MetaCity's early access period. Velathi is a fully developed constructed language with its own grammar, vocabulary, and cultural context — not a variation of any natural language, and not a language in which the 14 flagged words carry any offensive meaning. They are, by the assessment of community linguists who reviewed the list this morning, words roughly equivalent to 'morning,' 'together,' 'walk,' and 'return.' ClearFilter classified them as slurs.

The enforcement pattern reflects the breadth of the misclassification. The 11,000 accounts banned between midnight and 5:00 AM are not accounts that used the flagged words in aggressive or harassing contexts. They are accounts that used the flagged words in ordinary Velathi conversation — greetings, community posts, event announcements, casual exchanges. The ClearFilter system does not evaluate context; it evaluates vocabulary match. An account that posted a Velathi morning greeting using one of the 14 flagged words received the same permanent ban as an account that would have used an actual slur in an attack. The bans were issued silently, at night, to users who were asleep. Community members in the Velath cultural zones began waking up to ban notifications at approximately 5:00 AM and immediately recognized the pattern — the accounts being banned were all Velathi speakers, and the timing corresponded exactly to the ClearFilter model update.

ClearFilter Called Their Greetings Slurs. 11,000 Accounts Are Gone. No Appeals Accepted.

The appeal suspension is the dimension of the incident that has generated the most intense community response, and its logic — suspending appeals for accounts banned by a model that the platform has now confirmed is producing incorrect outputs — has been widely described as backwards. The standard position would be that a confirmed model error requires urgent account restoration, not a suspension of the mechanism that would allow affected users to seek it. MetaCity's stated reason for the appeal suspension is that it needs to review all of model version 14.2's outputs before processing individual appeals, to ensure that restorations are applied consistently. The practical effect for the 11,000 banned accounts is that they have no active path to reinstatement, no timeline for when one will exist, and no guarantee that the review will be completed before MetaCity's standard 30-day account termination process renders their bans permanent by default. The Velath cultural community has organized an emergency platform response. MetaCity has confirmed the error. No accounts have been restored.

The Bottom Line

No accounts have been restored.

You May Also Like