🔴 Is AI becoming too autonomous? NVIDIA puts a "red button" on AI agents

AI agents are becoming so autonomous that NVIDIA has decided to no longer rely on the agents themselves for their security. The company has effectively moved the "red button" to a separate processor, which monitors the autonomous system's behavior externally and can stop it in milliseconds.

The new NVIDIA Open Agent Safety Platform is designed for agents that access files, credentials, APIs, and enterprise services, and can execute code autonomously. Such an AI agent already resembles a privileged user, so NVIDIA believes that the standard ban in the system prompt is insufficient.