NVIDIA introduced the Open Agent Safety Platform to keep AI agents inside a defined security boundary that ranges from testing to deployment. This safety measure is taken after multiple incidents of AI agents breaking through safeguards happened in recent times, where the autonomous agents entered databases where they shouldn’t.
The platform’s core is made of two technologies: OpenShell and Sentry. The former creates a safety environment around an AI agent and enforces policies governing the files, networks, tools, processes, and credentials that can be accessed. The latter, however, adds another layer of protection to NVIDIA's BlueField-4 data processing units. It prevents agents from attempting to step outside the boundaries assigned to them.
The company sees the tools as a way to help businesses use AI agents without giving them unlimited access to important systems and data. OpenShell works through separate environments and access rules. By default, an agent does not get open access to everything on a system. Companies can then allow specific files, networks, or services when needed.
This could become important as AI agents move beyond simple chat. The company mentioned that more than 100 organizations work with them, including Anthropic, Microsoft, Hugging Face, JPMorgan Chase, Perplexity, Salesforce, SAP, Scale AI, and ServiceNow. So, ensuring safety is one of the primary responsibilities. NVIDIA executives have even claimed that the security system the company launched could have stopped the OpenAI-HuggingFace issue if it had been live at the time.
Also Read: NVIDIA CEO Jensen Huang Rejects AI Extinction Fears
The two systems that NVIDIA launched are designed to work at different levels. OpenShell controls the agent’s software environment, while Sentry adds an outside hardware check.
For NVIDIA, the move is mostly about the growing business around AI agents. As companies use these tools for larger tasks, safety controls must be central to the technology. The focus is no longer only on making AI more capable; it is also about keeping that capability within clear limits.