Cloudflare disclosed a cross-request KV cache exposure in the GLM-5.3 Workers AI model (@cf/zai-org/glm-5.3), active from August 27 to September 9, 2026, in which 155 responses to prompts longer than 681,280 tokens contained data from other customers' requests. The cause was a prefill KV pool smaller than the model's 1,048,576-token context window combined with an SGLang bug in disaggregated prefill/decode mode; the model was restored the same day with a corrected configuration (pool now ~2.04 million tokens) and patched engine, and no other Workers AI model was affected. Fixes landed in SGLa…
Read the full story on blog.cloudflare.com
What to doIf you run disaggregated SGLang, upgrade to v0.5.22 or later and check that your KV pool size exceeds your model's context length on every node.
- Provider
- Z.ai
- Source
- blog.cloudflare.com
- Models
- GLM-5.3glm-5.3