{ "version": 1, "event_id": "evt_83f75600ba2496e7", "url": "https://xiyu.news/events/evt_83f75600ba2496e7/", "json": "https://xiyu.news/api/events/evt_83f75600ba2496e7.json", "type": "security_incident", "status": "monitoring", "category": "technology", "title": { "zh": "GLM-5.3 Workers AI 模型中的跨请求 KV 缓存暴露（155 条响应）", "en": "Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)" }, "current_state": { "zh": "Cloudflare Workers AI GLM-5.3 的用户隔离泄露已修复，SGLang 相关缓存管理缺陷已修补并建议升级至 v0.5.22；未发现恶意利用，其他 Workers AI 模型未受影响。", "en": "The Cloudflare Workers AI GLM-5.3 user-isolation leak has been fixed; related SGLang cache-management defects were patched with a recommendation to upgrade to v0.5.22; no malicious exploitation was found and other Workers AI models were unaffected." }, "first_seen_at": "2026-10-27T09:57:58.995754+00:00", "last_updated_at": "2026-10-27T14:17:22.925649+00:00", "last_material_change_at": "2026-10-27T14:17:22.925649+00:00", "confidence": 0.75, "updates_count": 2, "sources_count": 2, "entities": [ "cloudflare", "cross", "glm", "gpu", "kv-cache", "workers" ], "identifiers": [ "glm-5" ], "topics": [ "ai-security", "cloudflare-workers-ai", "data-leak", "incident-postmortem", "kv-cache", "llm-inference", "multi-tenant-isolation", "sglang" ], "updates": [ { "update_id": "upd_a59159a58e66d666", "event_id": "evt_83f75600ba2496e7", "occurred_at": "2026-10-26T22:41:48Z", "published_at": "2026-10-26T22:41:48Z", "first_seen_at": "2026-10-27T09:57:58.995754Z", "time_precision": "published", "update_type": "initial", "material_change": true, "title_zh": "GLM-5.3 Workers AI 模型中的跨请求 KV 缓存暴露（155 条响应）", "title_en": "Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)", "what_changed_zh": "Cloudflare 发布了对 Workers AI 一个缺陷的根因分析：GLM-5.3 的跨请求 KV 缓存未在用户之间正确隔离，可能导致会话内存重叠，问题追溯至预填充缓存容量不匹配，已通过将该模型下线修复。", "what_changed_en": "Cloudflare published a root-cause analysis of a Workers AI bug in which GLM-5.3's cross-request KV cache was not properly isolated between users, allowing potential overlap of session memory, traced to a prefill cache capacity mismatch and remediated by pulling the model from serving.", "current_state_zh": "Cloudflare 发布了对 Workers AI 一个缺陷的根因分析：GLM-5.3 的跨请求 KV 缓存未在用户之间正确隔离，可能导致会话内存重叠，问题追溯至预填充缓存容量不匹配，已通过将该模型下线修复。", "current_state_en": "Cloudflare published a root-cause analysis of a Workers AI bug in which GLM-5.3's cross-request KV cache was not properly isolated between users, allowing potential overlap of session memory, traced to a prefill cache capacity mismatch and remediated by pulling the model from serving.", "detailed_summary_zh": "Cloudflare published a root-cause analysis of a Workers AI bug in which GLM-5.3's cross-request KV cache was not properly isolated between users, allowing potential overlap of session memory, traced to a prefill cache capacity mismatch and remediated by pulling the model from serving.", "detailed_summary_en": "Cloudflare published a root-cause analysis of a Workers AI bug in which GLM-5.3's cross-request KV cache was not properly isolated between users, allowing potential overlap of session memory, traced to a prefill cache capacity mismatch and remediated by pulling the model from serving.", "background_zh": "", "background_en": "", "community_discussion_zh": "", "community_discussion_en": "", "market_impact_zh": "", "market_impact_en": "", "importance_score": 8.0, "references": [ { "url": "https://news.ycombinator.com/item?id=50252159", "title": "Community discussion" } ], "confidence": 0.75, "story_ids": [ "hackernews:story:50252159" ], "sources": [ { "url": "https://blog.cloudflare.com/kv-cache-exposure-remediation/", "label": "wondersky", "source_type": "hackernews", "official": false } ] }, { "update_id": "upd_b5b41a602238852c", "event_id": "evt_83f75600ba2496e7", "occurred_at": "2026-10-27T10:12:02Z", "published_at": "2026-10-27T10:12:02Z", "first_seen_at": "2026-10-27T14:17:22.925649Z", "time_precision": "published", "update_type": "remediation", "material_change": true, "title_zh": "Cloudflare AI用户隔离失效：GPU缓存不够竟照常回答，155条回复泄露其他客户内容", "title_en": "Cloudflare AI用户隔离失效：GPU缓存不够竟照常回答，155条回复泄露其他客户内容", "what_changed_zh": "披露了另一类 SGLang 缓存重用漏洞、其修复版本 v0.5.22，以及实际泄露已于 9 月 9 日修复、未发现恶意利用且其他 Workers AI 模型未受影响。", "what_changed_en": "Disclosed a separate SGLang cache-reuse bug, its fix version v0.5.22, and that the actual leak was fixed on September 9, with no evidence of malicious exploitation and other Workers AI models unaffected.", "current_state_zh": "Cloudflare Workers AI GLM-5.3 的用户隔离泄露已修复，SGLang 相关缓存管理缺陷已修补并建议升级至 v0.5.22；未发现恶意利用，其他 Workers AI 模型未受影响。", "current_state_en": "The Cloudflare Workers AI GLM-5.3 user-isolation leak has been fixed; related SGLang cache-management defects were patched with a recommendation to upgrade to v0.5.22; no malicious exploitation was found and other Workers AI models were unaffected.", "detailed_summary_zh": "Cloudflare disclosed that a GPU cache capacity mismatch in its Workers AI GLM-5.3 deployment let 155 over-length requests leak other customers' cached content, alongside a related SGLang cache-reuse bug that the SGLang team has since patched.", "detailed_summary_en": "Cloudflare disclosed that a GPU cache capacity mismatch in its Workers AI GLM-5.3 deployment let 155 over-length requests leak other customers' cached content, alongside a related SGLang cache-reuse bug that the SGLang team has since patched.", "background_zh": "", "background_en": "", "community_discussion_zh": "", "community_discussion_en": "", "market_impact_zh": "", "market_impact_en": "", "importance_score": 7.5, "references": [], "confidence": 0.9, "story_ids": [ "telegram:theblockbeats:202574" ], "sources": [ { "url": "https://m.theblockbeats.info/flash/373940?from=telegram", "label": "theblockbeats", "source_type": "telegram", "official": false } ] } ] }