Cloudflare Says GLM-5.3 Bug Leaked 155 Responses Across Tenant Sessions

Cloudflare disclosed a security incident involving its GLM-5.3 model on Workers AI, where 155 responses to very long prompts leaked data across tenant sessions due to a KV cache misconfiguration and a bug in the SGLang inference engine. The issue was reported by a security researcher and fixed within two days, with no evidence of deliberate exploitation.