Cross-request KV cache exposure in the GLM-5.3 Workers AI model, and our remediation with SGLang
Factual evidence
What the source reports
Cloudflare resolved a Workers AI configuration error that triggered an SGLang bug, exposing cross-request KV cache data in GLM-5.3.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- Cloudflare Blog · ENTERPRISE AI
- Published by source
- 26 October 2026
- Collected by OneBench
- 27 Oct 2026, 08:01 UK
Stored source excerpt
A configuration mismatch in GLM 5.3 on Workers AI triggered an SGLang bug that exposed KV cache pages across requests. Here is how we addressed…
Short excerpt from the collected text, not the full source. Use the source link to read it in context.
The factual summary is a OneBench synthesis, not a quotation or independent verification. Collection time is not publication time. Open the source for its full context; related reporting can share the same underlying announcement.
OneBench interpretation
Institutional assessment
So what
Multi-tenant LLM inference infrastructure carries subtle memory-isolation risks that can leak sensitive prompts across distinct sessions.
Do what
Review vendor data-isolation guarantees with the team responsible for enterprise AI platform operations before using shared inference endpoints.