Source: The Hacker News
Researchers identified critical remote code execution vulnerabilities affecting major AI inference engines, including those developed by Meta, Nvidia, and Microsoft, alongside open-source projects like vLLM and SGLang. The common root cause for these vulnerabilities is the unsafe implementation of ZeroMQ (ZMQ) and Python’s pickle deserialization, which exposes systems to potential exploitation.
👉 Pročitaj original: The Hacker News
