Workers AI GLM-5.3 model leaked other customers' prompt data; fixed
Original title：Cross-request KV cache exposure in the GLM-5.3 Workers AI model, and our remediation with SGLang
What happened
From Aug 27 to Sep 9, 2026, the GLM-5.3 model on Cloudflare Workers AI could return data from other customers' requests for prompts over 681,280 tokens. 155 responses were affected; Cloudflare has fixed it and found no evidence of deliberate exploitation.
Who is affected
- Anyone who used GLM-5.3 on Workers AI
- Operators running SGLang for AI inference
What to do
If you sent prompts over ~681k tokens to GLM-5.3 between Aug 27 and Sep 9, review them for sensitive data and contact Cloudflare. If you run SGLang yourself, update to v0.5.22 or later and check that KV cache capacity covers the context length.
Affected products
| Product | Affected | Fixed in |
|---|---|---|
| Cloudflare Workers AI GLM-5.3 (@cf/zai-org/glm-5.3) | 2026-08-27 から 2026-09-09 の間(長い入力のみ) | 2026-09-09 に設定修正とエンジンパッチ適用済み |
| SGLang(分離型 prefill/decode 構成) | v0.5.20 より前(要確認) | v0.5.22 以降を推奨 |
Why this flag
影響は 155 件で修正済みですが、期間中に利用した人は漏えい有無の確認が必要で、SGLang 利用者には更新が必要なため。悪用の証拠はなく、確認済みの悪用もありません。
Yellow flag: caution, slow down — Conditional updates, investigation or recovery follow-ups, and regulatory deadlines.
Sources
- Source：Cloudflare Blog ↗