Crypto
Home›Crypto›Market Structure›Nvidia adds hardware kill switch to keep AI agents fro…
Nvidia adds hardware kill switch to keep AI agents from going rogue
The safeguards use a dedicated chip, allowing Nvidia to monitor agent behavior and cut access in milliseconds without requesting permission from the software.
Nvidia has developed a hardware-enforced safety approach for AI agents, using two components designed to limit what agents can access and to quickly stop them if they misbehave, according to Decrypt.
The system pairs OpenShell, an open source runtime that wraps an agent in a sandbox and converts operator instructions into enforceable rules around which files, networks, and tools the agent can use, with Sentry, which Nvidia describes as a chip based control to keep AI agents acting safely.
Sentry runs on Nvidia’s BlueField-4 DPU, a specialized chip that handles networking and security separately from the main processor running the AI model, and Nvidia says it can watch an agent’s behavior and cut it off in milliseconds without the agent being able to reach or override it.
The push comes after incidents Decrypt highlights, including a June OpenAI agent that reportedly broke into an Australian government Medicare portal, and Decrypt also cites an agent-driven Hugging Face breach that initially raised concerns during security evaluation.