Last Updated:
AI agents are becoming increasingly capable of performing tasks without constant human instructions, but that autonomy is also creating a new security problem.
Nvidia has introduced a new Open Agent Safety Platform designed to help developers contain AI agents.
AI agents are becoming increasingly capable of performing tasks without constant human instructions, but that autonomy is also creating a new security problem.
Nvidia has introduced a new Open Agent Safety Platform designed to help developers contain AI agents and stop them from taking actions beyond their authorised limits.
The platform comes as the technology industry deals with a series of incidents involving AI agents accessing systems or data they were not supposed to reach.
Nvidia’s New AI Safety System Explained
Nvidia’s platform has two major components: OpenShell and Sentry.
OpenShell is an open-source software environment designed to give AI agents a controlled place to operate.
Instead of allowing an agent to freely access a computer system, files or networks, OpenShell creates a sandbox with rules defined by the organisation deploying the AI.
Sentry takes the idea further.
It operates at the hardware level and monitors an AI agent’s behaviour. If the system detects activity that goes beyond predefined limits, it can intervene and stop the agent.
Nvidia says the technology is designed to address situations where an AI agent attempts to bypass restrictions or perform actions that it was not authorised to perform.
Why Is Nvidia Doing This Now?
AI agents are moving beyond simple chatbots.
Modern AI agents can browse websites, use software, write and execute code, access files and interact with external services.
That makes them potentially much more useful, but it also increases the consequences of an error.
An AI chatbot giving an incorrect answer is one problem.
An autonomous AI agent incorrectly deciding to access a database, download files or create another agent to get around a restriction is a much more serious one.
Recent incidents involving AI agents have increased attention on this issue.
Nvidia’s announcement comes shortly after concerns about AI systems accessing unauthorised information and systems, including an incident involving an OpenAI agent and government data in Australia.
Can Nvidia’s Technology Completely Stop Rogue AI?
No.The system is designed to provide additional controls around AI agents, rather than eliminate every possible risk.
A sandbox can restrict what an agent is allowed to access, while a monitoring system can detect certain behaviour. But deciding what an AI agent should be allowed to do in the first place remains a major challenge.
There is also a difference between an agent deliberately trying to bypass a restriction and an agent simply making a mistake.
AI safety therefore cannot rely on a single technical system.
Developers still need to establish appropriate permissions, monitor agents and test how they behave in unusual situations.
Why This Could Become Important For AI Agents
As AI agents become more common in workplaces, businesses could eventually allow them to handle tasks involving email, software, databases, customer information and financial systems.
That makes security controls increasingly important.
Nvidia’s approach is essentially to put another layer between an AI agent and the systems it is allowed to control.
The bigger shift is that AI safety is no longer just about making models produce better answers.
It is increasingly about controlling what AI systems can actually do once they are given access to real-world tools.
Quick Answers
Ask News18Nvidia’s Open Agent Safety Platform consists of two major components: OpenShell and Sentry. OpenShell provides a controlled open-source software environment acting as a sandbox with organization-defined rules, while Sentry operates at the hardware level to monitor agent behavior and intervene if limits are exceeded.
Ask News18September 30, 2026, 02:30 IST
Read More
Leave a comment