Cloudflare fixes GLM-5.3 cross-request cache leak
What happened
On October 27, Cloudflare disclosed that GLM-5.3 on Workers AI leaked KV cache data across requests, affecting 155 responses. The issue is fixed. All affected prompts exceeded 681,280 tokens. An undersized cache pool combined with a rejection-path bug in SGLang's disaggregated prefill/decode mode let responses read other users' leftover cache. Cloudflare says it found no evidence of deliberate exploitation and that other models were unaffected.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageCross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)
Cloudflare 披露，Workers AI 上的 GLM-5.3 出现跨请求 KV cache 数据泄露，共影响 155 条响应，问题已修复。受影响提示词均超过 681,280 tokens，缓存池容量不足与 SGLang 在分离式 prefill/decode 模式下的拒绝路径漏洞叠加，导致响应读取其他用户的残留缓存；未发现蓄意利用的证据，其他模型未受影响。
Heat over time
Not enough continuous observations to draw a trend yet.