Discord admits AI moderation bug wrongfully banned users over harmless images

Discord has publicly apologized for a major bug in its automated safety systems that wrongfully banned roughly 8,200 users over a two-month span starting in May 2026.

The issue triggered widespread backlash across social media after users discovered their accounts were permanently suspended for uploading completely harmless, everyday images.

1. The Catalyst: Square Grid Patterns

The automated safety system mistakenly flagged standard images containing square grid patterns as highly malicious content—specifically Child Sexual Abuse Material (CSAM). Because of how the automated filters were tuned, uploading any of the following benign items could trigger an immediate automated penalty:

• Video game screenshots (such as a player’s Minecraft inventory menu).
• Images of chessboards or complex spreadsheets.
• Digital game development textures.
• Standard graphics featuring checkerboard-style white and gray transparent backgrounds.

2. Anatomy of the Moderation Bug

According to an official technical breakdown released by Discord Support, the system failed across two distinct layers of the moderation pipeline:

Plaintext

[ THE DISCORD MODERATION BREAKDOWN ]

  1. THE FLAG (Similarity Matching)
  Automated system scans image ──► Flagged as false positive due to grid hash similarities
                                          │
                                          ▼ [ The Intended Safeguard ]
                                          │ Human review is queued; uploads paused
                                          │
  2. THE GLITCH (The Bug Takeover)
  System bypasses pause protocol ──► INSTANT PERMANENT ACCOUNT BAN ISSUED
                                          │
                                          ▼ [ The Final Roadblock ]
  3. THE REVIEW (Cleared but Stuck)
  Human staff clears the user     ──► Bug blocks automated unban; account stays suspended

• Bypassing the Upload Pause: The automated system uses “perceptual hashing” to match uploaded files against databases of known illicit material. When a similarity match occurs, it is supposed to temporarily pause the user’s ability to upload new files while a human member of the Trust & Safety team reviews the case. Instead, the bug caused the system to skip the pause step and issue an immediate, permanent ban.
• The Broken Unban Loop: Even more frustrating for users, when human moderators manually reviewed the flags and cleared the accounts of any wrongdoing, the exact same bug prevented the system from lifting the ban automatically, leaving innocent users locked out of their accounts indefinitely.

3. Current Status & Restoration

The problem was pushed to a boiling point when more than 200 users were caught in the false-positive trap over a single weekend, forcing the company to identify and deploy a backend patch.

Discord CTO and co-founder Stanislav Vishnevskiy confirmed that the software error has been resolved. The platform has officially reinstated all 8,200+ accounts caught up in the glitch. The company noted that it is actively overhauling its automated safety guardrails to ensure that false-positive flags do not result in quiet, unreviewable penalties moving forward.