Key facts
- Discord's AI moderation system incorrectly banned more than 8,000 users over two months.
- The bug flagged harmless images, such as spreadsheets, chessboards, and transparent backgrounds, as harmful content.
- Affected accounts are being restored, and Discord is working on improved safeguards.
- Users reported permanent suspensions for uploading images with grid patterns, speculated to be due to AI sensitivity to patterns used to disguise harmful content.
Discord has admitted that a bug in its artificial intelligence moderation system led to the wrongful banning of over 8,000 users in the past two months. The system incorrectly flagged harmless images, including spreadsheets, chessboards, game textures, and even simple white or gray transparent backgrounds, as violating its policies.
The company confirmed the issue began in May and that a fix was implemented after an additional 200 users were banned over a recent weekend. Discord is currently in the process of restoring all accounts affected by the bug.
This incident underscores the challenges platforms face as they increasingly depend on AI for content moderation at scale. Discord explained on X that its automated safety system matches uploaded content against databases of known harmful material, a process that can result in false positives. While human moderators typically review flagged content, the bug caused immediate bans.
Users on social media platforms like X and Reddit have shared experiences of being permanently suspended for uploading images with grid patterns. Some speculate that the AI became overly sensitive to such patterns due to their past use in attempts to disguise inappropriate content from detection systems.
Affected users have voiced significant frustration, highlighting how permanent bans based on automated detection can severely impact individuals who rely on Discord for professional communication, gaming communities, or social connections. Similar moderation issues have been reported by users of Instagram, Facebook Groups, and Tumblr in the past, with AI systems often suspected as the cause, though companies have not always publicly confirmed these errors.