Cloudflare's GLM-5.3 leak exposed other users' data in 155 AI responses
Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)
A misconfigured KV cache pool on Cloudflare's Workers AI deployment of GLM-5.3, combined with an SGLang bug in disaggregated prefill/decode mode, caused 155 responses to prompts over 681,280 tokens to include data from other customers' sessions. Cloudflare took the model offline within 40 minutes of declaring an incident, patched the engine, resized the pool to roughly 2.04 million tokens, and pushed fixes upstream to SGLang v0.5.22.
The decode node received a "transfer complete" message and assumed that prefill overwrote the KV cache from previous requests with the full prompt for this request, when in fact it left the majority of the previous state intact.