Summary: Cloudflare disclosed on October 26 that its GLM-5.3 model on Workers AI returned 155 responses containing data from other customers’ requests between August 27 and September 9, 2026. Only prompts longer than 681,280 tokens triggered the leak, caused by a settings mismatch plus a bug in the SGLang inference engine. The model was fixed and restored on September 9, no other Workers AI model was affected, and Cloudflare found no evidence anyone exploited it deliberately.
Storyline:
Hook: 155 AI answers contained another customer’s data.
Context: Cloudflare disclosed it October 26: between August 27 and September 9, its GLM-5.3 model on Workers AI mixed up customers. The model is fixed, and no other model was affected.
Stakes: Only giant prompts triggered it, over 681,280 tokens. A settings mismatch plus a software bug opened the gap, and Cloudflare found no sign anyone exploited it deliberately.
Payoff: 155 mixed-up answers prove one thing: check before you trust. Send this to the one person who pastes secrets into AI.
Watch next: UK consultation bots — 90,835 automated responses out of 118,756 to a 2023 tobacco and vapes consultation, a warning about who is really answering public surveys.
Links:
Instagram reel: https://www.instagram.com/p/DfDdCDNAFeM/
Sources: https://www.unite.ai/cloudflare-says-glm-5-3-bug-leaked-155-responses-across-tenant-sessions/ , https://aicybr.com/blog/cloudflare-glm-5-3-kv-cache-leak-sglang-remediation , https://bytenib.com/post/cloudflare-workers-ai-glm-5-3-cross-tenant-leak
https://www.instagram.com/simpleaiexplains/
https://buymeacoffee.com/simpleaiexplains
Leave a Reply