NCCapital

Nvidia is expanding beyond the chips and computing systems powering artificial intelligence with a new effort aimed at making autonomous AI agents safer. The company on Monday introduced the Open Agent Safety Platform, an open-source security framework designed to prevent AI agents from going off-task, accessing unauthorized resources or taking potentially harmful actions.

The platform arrives as AI companies increasingly shift their attention from chatbots that simply answer questions toward agents capable of independently completing multi-step tasks. These systems can browse websites, write and execute code, interact with applications and use external tools, giving them considerably more power—and creating new security risks when something goes wrong.

Nvidia’s Open Agent Safety Platform is built around two open-source technologies called OpenShell and Sentry. OpenShell provides an isolated environment where agents can operate under defined policies, while Sentry is designed to monitor their behavior and identify potentially dangerous activity before it spreads beyond that controlled environment.

Justin Boitano, who leads Nvidia’s enterprise AI efforts, said the system can “quarantine a suspicious agent in milliseconds.” The objective is to give developers and companies a security layer capable of responding immediately when an autonomous system begins behaving in an unexpected or unauthorized way.

OpenShell effectively places boundaries around what an AI agent is allowed to do. Developers can determine which files, networks, tools and computing resources an agent can access, helping prevent a compromised or malfunctioning system from freely interacting with the broader infrastructure around it.

Sentry adds another layer by evaluating agent behavior for suspicious activity. Together, the tools are intended to provide companies with controls that become increasingly important as businesses allow AI systems to perform real work rather than simply generate text or images.

CEO Jensen Huang described the launch as the beginning of an open ecosystem for creating a “trust layer” around agentic AI. Nvidia’s decision to make the technology open source is particularly notable, allowing developers and security researchers to inspect, modify and build on the platform rather than relying exclusively on a proprietary security system.

The initiative also represents another expansion of Nvidia’s ambitions beyond hardware. The company already dominates the market for accelerators used to train and run advanced AI models, but it has increasingly invested in software, networking, inference systems and enterprise AI tools that surround its processors.

Agent security could become an increasingly important part of that ecosystem. As autonomous AI systems gain access to corporate databases, internal software, payment systems and other sensitive resources, enterprises will need ways to restrict what those agents can access and quickly contain them when their behavior deviates from established policies.

Nvidia made another major financial announcement Monday, with its board authorizing an additional $150 billion for the company’s share repurchase program. The expansion is the largest new buyback authorization announced by a U.S. company, underscoring the enormous amount of cash being generated by the AI boom.

The authorization gives Nvidia significantly more flexibility to return capital to shareholders while continuing to invest heavily in research, new processors, networking technology and AI infrastructure. The scale of the buyback also reflects how dramatically Nvidia’s financial position has changed as demand for AI computing has transformed it into one of the world’s most valuable companies.

Taken together, Monday’s announcements highlight the two sides of Nvidia’s current position. The company is generating enough financial firepower to authorize a record-setting $150 billion expansion of its buyback program while simultaneously pushing into emerging areas such as autonomous-agent security. As AI evolves from models that generate answers into systems capable of taking actions on their own, Nvidia is betting that securing those agents could become another critical layer of the AI technology stack.