AI Security Threat Feed

vLLM: uninitialized tensor lets tenant crash engine

vLLM: num_frames override enables memory-exhaustion DoS

vLLM: multimodal metadata DoS crashes inference engine

vLLM: decompression bomb in Nemotron-VL causes DoS

vLLM: SSRF via X-Request-Id leaks NCCL KV-cache data

vLLM: gRPC invalid token IDs crash EngineCore (DoS)

vLLM: gRPC logprob limit bypass crashes EngineCore

vLLM: gRPC top_n validation flaw crashes EngineCore

virtualenv: GBK quote bypass in xonsh activation enables RCE

virtualenv: prompt injection enables shell command execution

filelock: SQLite symlink redirect enables file overwrite

Strands Agents SDK: A2A session leak across callers

virtualenv: bash prompt injection allows command execution

SiYuan: WebSocket auth bypass leaks live event stream

HF Datasets: pickle deserialization RCE via pandas builder

obsidian-tc: HITL approval bypass runs destructive tools

PraisonAI: unauthenticated A2U event streaming exposed

PraisonAI: sandbox fails open, bypassing FS/network limits

PraisonAI: recipe policy bypass self-approves shell tool

PraisonAI: missing ownership check lets members delete data