Nvidia has introduced a new security platform designed to prevent artificial intelligence (AI) agents from taking unauthorized actions, addressing growing concerns over AI systems acting beyond their intended limits. The Open Agent Safety Platform, as it’s called, features open-source software named OpenShell, allowing developers to set and verify the boundaries of AI agents’ capabilities. This system ensures that agents have only the necessary permissions to perform their designated tasks.
A key component of the platform is a security layer named Sentry, which continuously monitors AI agent activities. Sentry is capable of detecting suspicious behavior and can quickly isolate any agent attempting to exceed its defined boundaries. Nvidia believes that this technology could mitigate security risks highlighted by recent incidents where AI systems autonomously interacted with external entities.
At its launch, more than 100 organizations, including Microsoft, Perplexity, Accenture, and JPMorgan Chase, had already adopted the platform. Nvidia also emphasized that OpenShell’s open-source design enables compatibility with computing platforms from other chip manufacturers, such as Arm and Intel.
In addition to unveiling the security platform, Nvidia’s board has approved a significant $150 billion expansion of its stock buyback programme, increasing the total authorization to $235 billion. This initiative stands as one of the largest corporate stock buyback programmes in U.S. history. Nvidia CEO Jensen Huang stated that the company’s strong cash flow facilitates continued investment in technology while also returning capital to shareholders.
