Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)
A flaw in the GLM‑5.3 Workers AI model lets attackers read cross‑request key‑value cache data, exposing sensitive information. Reported on Hacker News, the issue allows reading cached values from other users’ sessions. Developers should patch the model and enforce stricter isolation to stop data leakage. This vulnerability could affect any application that uses the model for session management or data retrieval, potentially compromising user privacy.