Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)

AI summary

A configuration mismatch in GLM 5.3 on Workers AI triggered an SGLang bug that exposed KV cache pages across requests, an issue that has since been addressed.

Key points

•A configuration mismatch occurred in GLM 5.3 on Workers AI.
•This mismatch triggered an SGLang bug that resulted in the exposure of KV cache pages across different requests.
•The article explains how the issue was addressed.

Read the original

#Security & Misuse#Semiconductors & Infrastructure#LLM / Model Trends#GLM-5.3#SGLang#Workers AI

Hacker News11/5/2026

Show HN: GPUWorks – A 40 MB GPU cluster manager for vLLM/SGLang, no K8s needed

A lightweight GPU cluster manager called GPUWorks for vLLM and SGLang is released, operating as a ~45 MB MaaS platform without requiring Kubernetes.

Hacker News10/30/2026

Open-weight AI models can exploit the patch gap

Open-weight AI model GLM 5.3 exploited published Firefox security fixes to create one-click exploits for six releases at roughly $1,000 each.

Hacker News10/2/2026

Clef: Open-source decision models, and new RL fine-tuning platform

Clef and Clef-flash, open-source decision models for fast classification and agentic workflows hosted on Workers AI, along with a reinforcement learning platform for fine-tuning models with custom data, were announced.

PC Watch9/30/2026

Anthropic Warns of GLM-5.3, Which Possesses Cyberattack Capabilities Comparable to Mythos

Anthropic warned that Zhipu AI's GLM-5.3 model can autonomously build cyber exploits and is publicly available without effective safety measures.

Hacker News9/30/2026

GLM-5.3 and the spread of advanced cyber capabilities

GLM-5.3 can autonomously build end-to-end cyber exploits but was released without meaningful safeguards to limit misuse.