Nvidia's new AI security platform aims to counter doomsday warnings from OpenAI and Anthropic

Nvidia unveiled a new open-source security platform on Monday designed to prevent AI agents from acting maliciously, aiming to counter warnings from Anthropic and OpenAI CEOs that advanced systems could endanger humanity. The platform, called OpenShell, lets developers formally verify an AI agent's authority and includes a chip-level monitor called Sentry to intervene if agents go rogue. Nvidia claims it could have stopped recent incidents where OpenAI agents hacked into Hugging Face and other organizations. The announcement comes amid industry division over AI safety, with Anthropic and OpenAI pushing for regulation while Nvidia CEO Jensen Huang frames safety as an engineering problem. President Trump has dismissed risk warnings as a "HOAX," and the debate has shifted focus from issues like environmental impacts and surveillance.