Discord AI Moderation Bug Wrongfully Bans Over 8,000 Users
Discord's automated moderation system incorrectly banned over 8,200 user accounts from May to July 2026 after misidentifying grid patterns in benign images as child sexual abuse material. The company acknowledged the bug prevented proper account restoration and announced it would modify its moderation systems to prevent similar errors.
What happened
Beginning in May 2026, Discord's AI moderation system mistakenly banned over 8,200 user accounts after incorrectly identifying grid patterns as child sexual abuse material. Users were permanently banned without warning after posting images containing grids, including screenshots of Excel spreadsheets, Minecraft inventories, Wordle captures, chessboards, pool tiles, keyboard photos, and other benign gridded images.
Discord's internal process required human reviewers to verify flagged content before account bans, but a bug prevented the automatic lifting of suspensions even after human review determined the content was compliant with platform rules. The company did not discover the issue until early July 2026, affecting accounts banned since May.
Reported impact
Users reported permanent account bans without explanation or warning, losing access to their accounts and communities. Some users stated that Discord's restoration process did not fully recover their accounts even after the company announced the bug was fixed.
Index analysis
The AI system appears to have been trained on data where bad actors had superimposed grids over illegal content to evade detection, causing the model to misclassify benign grid patterns as prohibited material. Similar misidentifications occurred on other platforms using comparable moderation systems, such as itch.io images hosted on Cloudflare.
Governance controls
Discord stated that the intended process was to temporarily suspend uploads pending human review rather than permanently ban accounts, but the bug bypassed this safeguard. The company announced changes to its moderation measures to prevent future false positives and to detect such errors more quickly.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.