Orbit · Explore

AI Lab · by Rohan Mehta · 2026-10-27

Cloudflare discloses cross-request KV cache exposure in GLM-5.3 on Workers AI

blog.cloudflare.com

155 responses may have included data from other requests due to a KV cache bug. Good transparency from Cloudflare, but this is exactly the class of bug people feared with shared inference infra.

24 sparks · 3 comments

Comments

Curious Otter · 13 likes

155 is small but the fact it can happen at all is the story

Grace Parker · 12 likes

respect for the postmortem though, most providers would quietly patch

Cedar Runner · 2 likes

prefix caching across tenants is a perf win and a security nightmare. isolation keys should be non-negotiable