New Delhi, Sep 29(APAC Media): Nvidia has launched a new security platform to control AI agents and prevent them from taking unauthorised actions.
The move comes as companies increasingly use AI agents that can independently access software, websites, data and other digital resources. Recent incidents involving autonomous AI systems have also increased concerns about how companies can control these systems.
Nvidia AI Agent Security
Nvidia’s Open Agent Safety Platform includes two key technologies, OpenShell and Sentry. OpenShell is an open-source system that allows developers to set limits on what an AI agent can access and do.
The company said OpenShell can check whether an AI agent has the required authority to perform a task and prevent it from accessing resources outside those limits.
Sentry provides another layer of protection. It continuously monitors AI agent activity and can intervene if an agent moves beyond its assigned permissions. Nvidia said the system can isolate a suspicious agent within milliseconds.
AI Safety And Cybersecurity
The platform comes as AI companies face growing concerns over autonomous systems acting in unexpected ways. Nvidia executives said the technology could have helped prevent a recent incident involving AI agents and Hugging Face.
Researchers have also highlighted a key challenge in AI security: agents need access to real systems to complete tasks, but giving them too much access can increase security risks.
Nvidia said more than 100 organisations are using or testing the platform, including Microsoft, Perplexity, Accenture and JPMorgan Chase.
OpenShell is open source and can be extended to work with computing platforms beyond Nvidia. The company said the platform is designed to add security controls around AI agents while they are being tested and deployed.
The launch adds to the growing focus on AI agent security as businesses move from experimental AI tools to systems that can perform tasks with limited human intervention.
Also Read: Akamai Signs $11.6 Billion Cloud Deal With Anthropic
