Cloudflare has addressed a security vulnerability in its GLM-5.3 Workers AI model that could expose data between customer requests. The issue, reported by Gadi Evron of Knostic, occurred between August 27 and September 9, 2026, affecting a small number of responses to prompts exceeding 681,280 tokens. The vulnerability stemmed from a misconfiguration in the SGLang inference engine, where the KV cache pool was smaller than the model's context window, leading to stale data being included in responses. Cloudflare has since corrected the configuration and patched the inference engine, restoring the model's service. AI
IMPACT Highlights the ongoing need for robust security measures and careful configuration management in AI model deployments, especially those handling long contexts.
RANK_REASON Security vulnerability and remediation for a specific AI model deployment.
Read on Hacker News — AI stories ≥50 points →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
