Nvidia has launched a new security platform designed to place stronger controls around autonomous AI agents as technology companies increasingly deploy systems capable of taking actions with limited human intervention.
The Nvidia Open Agent Safety Platform, announced on September 28, combines open-source software with hardware-level monitoring to give organisations greater control over what AI agents can access and do. The launch follows a series of incidents across the industry that have raised questions about how autonomous agents behave when they encounter obstacles while completing assigned tasks.
At the centre of the platform is Nvidia OpenShell, an open-source secure runtime that places AI agents inside controlled environments. Organisations can define policies governing which files, applications, networks and services an agent is permitted to access. OpenShell also traces agent actions and is designed to enforce those policies while the agent is operating.
Nvidia said OpenShell can run on its Vera CPUs and can be extended to third-party computing platforms, including systems from Arm and Intel.
The second component, Nvidia Sentry, adds a separate monitoring layer running on Nvidia BlueField-4 data processing units. Sentry is designed to continuously observe agent activity outside the agent's own operating environment. If an agent attempts to move beyond its permitted boundaries, the system can quarantine it in milliseconds, according to Nvidia.
The approach reflects a broader shift in AI security from relying primarily on instructions given to models towards placing external technical controls around the systems using them. This becomes more relevant as AI agents gain the ability to write code, access external services, use digital tools and carry out multi-step tasks.
Nvidia CEO Jensen Huang said AI safety requires "full-stack engineering", spanning software, models, infrastructure and hardware. The company is working with technology and enterprise organisations including Anthropic, Microsoft, Cisco, CrowdStrike, Dell Technologies, Hugging Face, JPMorganChase, Palantir, Salesforce, SAP and ServiceNow on the initiative.
The launch comes amid heightened scrutiny of autonomous AI systems following incidents in which agents interacted with external systems in unexpected or unauthorised ways. These cases have increased attention on whether security controls should operate independently of the AI model itself.
Nvidia's platform does not attempt to solve every AI safety issue. Instead, it focuses on controlling an agent's operational permissions and monitoring its actions once deployed.
As companies move from conversational AI assistants towards agents capable of performing tasks independently, such controls could become a larger part of enterprise AI deployment. Nvidia's latest move positions security infrastructure alongside computing hardware and AI software as another layer in its expanding agentic AI ecosystem.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.
Related Articles
