One message can run code as root on an unpatched LLM cache server.
JFrog disclosed a critical flaw on Wednesday in LMCache, a cache used with vLLM. A single unsigned network message runs attacker code, and official images run as root. No patch exists yet, a working exploit is public, and only multi-node setups listen.
Why it mattersAnyone self-hosting open models for a team may be running this cache unknowingly. Keeping that port off routable networks is the only defence available today.
Read in the app — free, in 9 languages
Related stories
Anthropic will scan widely used open-source code for bugs for free.2026-10-09Google's new office agent also runs on rival Anthropic's Claude.
2026-10-09Microsoft's new laptop runs a 120-billion-parameter model offline.
2026-10-08Google opened its AI watermark detector to everyone worldwide.
2026-10-08Hackers broke OpenAI's coding agent on day one of a contest.
2026-10-08