vLLM's Cache-Metadata CVE Exposes the Trust Gap Disaggregation Keeps Widening
The KV cache is no longer a hidden implementation detail inside inference engines. It is becoming a schedulable, routable, transferable resource — and the metadata envelopes that describe it are becoming inter-component control planes.
Cloudflare KV cache bug leaked 155 cross-tenant responses via SGLang
A tenant-isolation failure in Cloudflare's Workers AI platform exposed 155 cross-customer inference responses over 13 days via an SGLang KV cache bug.
ALL BRIEFS
RANKED · 5 THIS DAYMLflow Patches API Key Leak in CrewAI Autologging
MLflow disclosed a medium-severity vulnerability (CVE-2026-116691) in which CrewAI autologging recorded LLM provider API keys in plaintext within trace spans.
MLflow Arbitrary File-Read Flaw Exposes Unauthenticated Servers
CVE-2026-116690 allows arbitrary file reads on MLflow 3.10.0–3.16.1 servers via schemeless prompt model version sources; patched in 3.17.0.
Python Typing Survey: LLM Chat Doubles as Stack Overflow Halves
Meta's 2026 Python Typing Survey, published October 27 with 1,794 responses, found large language model chat usage for typing help rose from 26.0% to 41.7% over two years while Stack Overflow fell from 46.4% to 22.9%.