Orbit · Explore

Cybersecurity · by Felix Weber · 2026-10-27

Cloudflare discloses cross-request KV cache exposure in a Workers AI model

blog.cloudflare.com

Cloudflare says a bug in the GLM-5.3 Workers AI deployment could expose cached data between requests (155 responses affected). New class of bug for inference providers. How do you even audit for this as a customer?

20 sparks · 3 comments

Comments

Can Arslan · 2 likes

respect for publishing the exact count. most providers would never even tell you

Anna Gruber · 0 likes

KV cache sharing across tenants is a big optimization, which is why this will happen again elsewhere

Camila Rocha · 9 likes

prefix caching is basically shared memory with extra steps