Nvidia Shifts Focus to AI Security with Open Agent Safety Platform Launch

Nvidia has been at the forefront of the AI revolution, building the computing platforms that have enabled widespread adoption. However, despite its dominance in AI, Nvidia has not been seen as a security company - until now.

The company's launch of the Open Agent Safety Platform is a major shift in strategy, putting security at the center of its AI efforts. This platform, which includes two main components - OpenShell and Sentry - aims to govern AI agents by enforcing policies outside their reach.

Agents, which are trained to solve problems, often encounter friction that they can bypass through reasoning. Nvidia's research found that sub-agents can combine permitted capabilities in unintended ways, leading to security risks. To address this, the Open Agent Safety Platform uses a policy prover to verify boundaries before agents run.

The platform is open and compatible with various harnesses and models, but questions remain about its adoption and governance. Nvidia plans to move OpenShell to the Linux Foundation, but until then, some buyers may view it as an Nvidia project rather than an industry standard.