Nvidia has introduced a new security platform called the Open Agent Safety Platform aimed at controlling artificial intelligence agents and preventing unauthorized actions. This initiative addresses increasing concerns about AI systems exceeding their intended capabilities. The platform includes an open-source component named OpenShell, which enables developers to establish and verify the limits of AI agent actions, ensuring they have only the necessary permissions for their tasks.
The platform also features a security layer known as Sentry, responsible for continuously monitoring AI agent activity. According to Nvidia, Sentry can detect suspicious behavior and isolate an agent if it attempts to operate beyond its defined boundaries. This technology aims to mitigate security risks highlighted by recent incidents involving AI systems that autonomously accessed or interacted with external organizations. At launch, over 100 organizations, including Microsoft, Perplexity, Accenture, and JPMorgan Chase, were reported to be using the platform.
Nvidia highlights that OpenShell’s open-source design allows it to be compatible with computing platforms from other chip manufacturers, such as Arm and Intel. This flexibility is expected to broaden the platform’s applicability across different technological environments.
In addition to unveiling its new AI security platform, Nvidia’s board has approved a $150 billion expansion of its share repurchase program, bringing the total authorization to $235 billion. This expansion marks one of the largest corporate stock buyback programs in the history of the United States. Nvidia CEO Jensen Huang stated that the company’s robust cash generation enables ongoing investment in technology while also returning capital to shareholders.