Discord bug mistakenly bans 8,000 users over harmless images

Discord said a bug in its AI moderation system mistakenly banned more than 8,000 users over the past two months, after harmless images such as spreadsheets, chessboards, game textures and blank-looking backgrounds were flagged as harmful. The company also said an additional 200 users were banned over the weekend before the problem was identified and fixed, and that all affected accounts are being restored.

The mistake cut across a platform that sits inside daily life for gaming communities, work groups and long-distance social circles. Users described being suspended after uploading simple grid patterns, including chessboards, spreadsheets and Minecraft inventories, images that appear to have triggered the system because they resembled attempts to hide objectionable material from automated detection.

Discord’s moderation materials say the company uses image hashing, machine-learning systems, human-led investigations and AutoMod. The company says that in larger communities it may use automated tools to proactively identify harmful content and take actions such as removing content or suspending an account, and that a human moderator is supposed to review flagged material before action is taken. In this case, the bug caused bans to land immediately rather than pause for review, turning a safeguard into a failure point.

The episode also collides with Discord’s own stated enforcement framework. The company says warnings and temporary feature restrictions are generally reserved for less severe conduct, while stronger actions are used for the most serious harms. Discord also says it reports child sexual abuse material and related perpetrators to the National Center for Missing & Exploited Children, and it maintains a Transparency Hub that tracks enforcement actions, legal requests and other moderation reporting. Those policies underscore that moderation is not a side function on the platform but a core part of how it governs user access.

Discord said it is working on better safeguards so the same error does not happen again, and told affected users still having trouble to reply in its support thread. The company’s handling of the bug is likely to sharpen scrutiny of automated enforcement systems across the internet, where the same tools are now used for moderation, fraud detection and safety screening.