A critical vulnerability within the multiprocess mode of LMCache, an open-source tool used to accelerate LLM servers, allows unauthenticated attackers to execute remote code. The flaw stems from insecure handling in the ZeroMQ messaging layer. At present, no official patched version has been released, requiring administrators running this AI infrastructure to implement manual network restrictions immediately.
READ THE ORIGINAL ARTICLE AT THEHACKERNEWS.COM →