Researchers Find Serious AI Bugs Exposing Meta, Nvidia, and Microsoft Inference Frameworks

Cybersecurity researchers have identified critical remote code execution vulnerabilities in prominent AI inference engines from Meta, Nvidia, Microsoft, and open-source projects like PyTorch's vLLM and SGLang. These vulnerabilities stem from the unsafe handling of ZeroMQ (ZMQ) messaging and Python's pickle deserialization, highlighting significant security risks in AI deployment frameworks that rely on these components.