Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)
Cloudflare disclosed that a misconfiguration plus an SGLang bug let GLM-5.3 on Workers AI return other customers' data in 155 responses to prompts longer than 681,280 tokens between August 27 and September 9. The company says it found no evidence of deliberate exploitation and has fixed both the configuration and the engine. The case matters because it exposes a tenant-isolation risk class in shared inference engines that other SGLang operators must now patch against.
SOURCES
- Hacker News (AI)2026-10-26 22:41