Nvidia has introduced a new security platform aimed at enhancing control over artificial intelligence agents, addressing concerns about AI systems operating beyond their intended boundaries. The Open Agent Safety Platform, which includes open-source software called OpenShell, allows developers to define and verify the limits of what an AI agent is authorized to do. This initiative is part of Nvidia’s effort to mitigate security risks associated with AI systems taking unauthorized actions.
A key feature of the platform is Sentry, a security layer that continuously monitors AI agent activity. Sentry can detect and isolate suspicious behavior, ensuring that AI agents stay within their defined operational limits. Nvidia highlights that this technology could help manage security challenges highlighted by recent incidents where AI systems interacted with external entities autonomously.
The platform’s design allows OpenShell to be extended to other computing platforms, including those from Arm and Intel, due to its open-source nature. This flexibility makes it accessible to a wide range of organizations, with over 100 companies, such as Microsoft, Perplexity, Accenture, and JPMorgan Chase, reportedly using the platform at its launch.
In addition to this security initiative, Nvidia’s board has approved a significant expansion of its stock buyback program, authorizing an additional $150 billion. This brings the total authorization to $235 billion, marking one of the largest corporate stock buyback programs in U.S. history. CEO Jensen Huang noted that the company’s robust cash flow enables continued investment in technology while also allowing capital returns to shareholders.