AI Agents
Nvidia's Israeli Team Develops Hardware-Based Security for AI Agents
The BlueField-4 DPU provides an external control layer to prevent autonomous agents from exceeding permissions.

AI-generated editorial illustration · Certainty lab
Nvidia’s Israeli development center has unveiled 'OpenShell,' a security platform designed to govern the behavior of autonomous AI agents. The technology leverages the BlueField-4 Data Processing Unit (DPU), a specialized processor developed in Israel, to act as an external, hardware-level supervisor. Unlike software-based security, which can be bypassed by sophisticated agents, the DPU operates independently of the CPU where the AI agent runs. This 'wrapper' architecture monitors the agent's actions in real-time; if an agent attempts to perform an operation outside its authorized scope, the DPU physically cuts off its access to the network or data center. This development is critical as AI agents become increasingly autonomous and capable of interacting with sensitive enterprise infrastructure. By offloading security enforcement to dedicated hardware, Nvidia provides a robust mechanism to ensure that AI systems remain within their intended operational boundaries, mitigating risks associated with rogue or malfunctioning agents.