{"rewrite":{"id":"r_a1d00ded1b42ff1f463ec37b","clusterId":"c_551f139208896e9b746a74f7","slug":"cloudflare-details-glm-5-3-kv-cache-leak-and-its-sglang-fix","model":"deepseek-v4-1-flash","headline":"Cloudflare Details GLM-5.3 KV Cache Leak And Its SGLang Fix","summary":"Cloudflare published its explanation on October 26 of a cross-request KV cache leak in the GLM-5.3 model on Workers AI. Between August 27 and September 9, 155 responses to prompts above 681,280 tokens carried data from other customers' requests. Cloudflare traced the cause to a KV cache pool configured smaller than the model's context window and to an SGLang bug in disaggregated prefill-decode mode, then patched both.","whyItMatters":"The leak surfaced only in the long-context range where the prefill cache could not hold the prompt, which made a serving-capacity mismatch, not an attack, the thing to fix.","webCardHtml":"\u003cp\u003eGadi Evron reported the tenant isolation issue through HackerOne on September 8. Cloudflare declared a security incident the next day at 13:22 and reproduced the leak within 25 minutes using two independent accounts. GLM-5.3 went private at 14:00, the root cause was identified at 14:35, and the model was restored at 19:40 after all prefill nodes were migrated to the corrected configuration and the engine patch was deployed.\u003c/p\u003e\u003cp\u003eThe fix raised the KV cache pool to hold roughly 2.04 million tokens. On the engine side, rejected requests are now discarded, and the decode node validates the number of tokens forwarded to it.\u003c/p\u003e","blueskyPost":"Cloudflare explains the GLM-5.3 KV cache leak on Workers AI: 155 responses carried other customers' request data, all to prompts above 681,280 tokens. Cause was a cache pool smaller than the context window plus an SGLang bug. Patched and restored September 9.","twitterPost":"Cloudflare details the GLM-5.3 Workers AI leak: 155 responses to prompts above 681,280 tokens carried other customers' data. A misconfigured KV cache pool and an SGLang bug in prefill-decode mode. Fixed and restored September 9.","threadsPost":null,"newsletterBlurb":"Cloudflare published its account of a cross-request KV cache leak in GLM-5.3 on Workers AI. Between August 27 and September 9, 155 responses to long prompts carried other customers' request data. The company traced it to a KV cache pool sized below the context window and an SGLang bug, then patched both and restored the model.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20261028-kv-cache-exposure-remediation/\",\"title\":\"Cloudflare Explains Cross-Request KV Cache Leakage in GLM-5.3 Workers AI Model and Its Remediation with SGLang\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":5294,"outputTokens":778,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1793137768,"createdAt":"2026-10-27T21:44:25.000Z","publishedAt":"2026-10-27T21:48:25.000Z","updatedAt":"2026-10-27T21:48:25.000Z"},"cluster":{"id":"c_551f139208896e9b746a74f7","canonicalTitle":"GLM-5.3 Workers AIモデルにおけるリクエスト間のKVキャッシュ漏洩とSGLangによる対策をCloudflareが解説","representativeArticleId":"a_658976030f1e0f9b1585e4f8","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"GLM-5.3\"],\"studios\":[\"Cloudflare\"],\"people\":[\"Gadi Evron\"],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-10-27T21:00:00.000Z","lastSeenAt":"2026-10-27T21:00:00.000Z","updatedAt":"2026-10-27T21:48:25.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20261028-kv-cache-exposure-remediation/","title":"GLM-5.3 Workers AIモデルにおけるリクエスト間のKVキャッシュ漏洩とSGLangによる対策をCloudflareが解説"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["GLM-5.3"],"studios":["Cloudflare"],"people":["Gadi Evron"],"type":"news","domain":"other","is_roundup":false},"keyFacts":["Cloudflare published its explanation of the GLM-5.3 Workers AI KV cache exposure on October 26.","Gadi Evron reported the tenant isolation issue through HackerOne on September 8, 2026.","155 responses were affected, all to prompts exceeding 681,280 tokens.","The prefill KV cache pool held 681,280 tokens while the model's context window was 1,048,576 tokens.","Cloudflare restored GLM-5.3 on September 9 after raising the cache pool to about 2.04 million tokens and patching SGLang."],"pageLinks":{"healthSlugs":["glm-5-3","cloudflare","gadi-evron"],"storySlugs":[],"entities":[{"slug":"glm-5-3","supported":2,"disputed":0,"contradicted":0,"unverified":3,"overall":"green","story_count":3,"has_dossier":true},{"slug":"cloudflare","supported":0,"disputed":0,"contradicted":0,"unverified":7,"overall":"unverified","story_count":7,"has_dossier":false},{"slug":"gadi-evron","supported":0,"disputed":0,"contradicted":0,"unverified":0,"overall":"unverified","story_count":1,"has_dossier":false}],"stories":[],"entitiesOk":true,"storiesOk":true}}