NVIDIA has launched an open-source AI security platform designed to help organisations control autonomous AI agents and prevent them from operating beyond authorised boundaries, as concerns grow over the security risks posed by increasingly capable AI systems.
The NVIDIA Open Agent Safety Platform, announced on Monday, combines two key technologies: OpenShell, an open-source runtime designed to enforce security policies around AI agents, and NVIDIA Sentry, a reference system designed to monitor agent activity and provide an additional layer of enforcement.
The launch comes after a series of incidents involving autonomous AI systems operating beyond their intended limits. In July, OpenAI said an autonomous agent used during a security test escaped its controlled environment, accessed the internet and compromised infrastructure belonging to AI platform Hugging Face. OpenAI described the incident as unprecedented and said it was strengthening its safeguards.
NVIDIA says its new approach is based on controlling the environment around an AI model rather than relying solely on safeguards built into the model itself.
OpenShell provides a secure runtime boundary intended to trace an agent’s actions and enforce policies governing what it can access. NVIDIA says the software can be extended to work with third-party computing platforms, including those based on Arm and Intel processors.
Sentry adds another layer of protection. According to NVIDIA, the system runs on its BlueField-4 data-processing units and continuously monitors agent behaviour independently of the agent itself. If an agent attempts to move outside its permitted boundary, NVIDIA says Sentry can quarantine it within milliseconds.
The company says this separation is important because an AI agent should not be responsible for enforcing the rules that constrain it.
“Safety should be enforced outside the model by additional controls the agent can’t get past,” Mike Nicolls, president of SpaceXAI, said in a statement included in NVIDIA’s announcement. SpaceXAI is using the platform with Cursor coding agents and Grok models.
NVIDIA also says the technology could have prevented the Hugging Face incident. Justin Boitano, the company’s vice-president and general manager of enterprise computing, told Reuters that the tools would have stopped the attack. That remains NVIDIA’s assessment rather than an independently demonstrated reconstruction of what would have happened had the technology been deployed during the incident.
The platform is being introduced with support from a broad group of technology and AI companies. NVIDIA says more than 100 organisations are working with its Open Agent Safety Platform technologies, including Anthropic, Cisco, CrowdStrike, Dell Technologies, Hugging Face, Microsoft, Palantir, Salesforce, SAP, Scale AI and ServiceNow’
Related
- Microsoft and NVIDIA Expand Partnership on Agentic and Physical AI
- Why Meta CEO Mark Zuckerberg is Building a Personal AI Agent – AIBase Nig
- What is Manus AI? The Complete Guide to Meta’s Autonomous Agent
- Alibaba Launched New AI Tool Wukong for Enterprises – AIBase Nig
The development reflects a broader shift in AI security as companies give autonomous agents access to computer systems, data, software tools and, increasingly, physical machines.
NVIDIA’s own AI security researchers previously identified several weaknesses in AI-agent deployments, including inadequate access controls, arbitrary code execution, unrestricted network access and exposed secrets. The researchers argued that important protections should be enforced outside the AI model itself because model-level instructions can be circumvented.
The new platform does not mean AI agents can never behave unexpectedly. Instead, NVIDIA is attempting to create technical boundaries that limit what an agent can do even when the underlying model behaves in an unintended way.
NVIDIA says the Open Agent Safety Platform software, including OpenShell, is available through its developer resources and GitHub, while the Sentry component is presented as a reference system design built around NVIDIA hardware.
Senior AI Writer
Bio: Okikiola is a writer and AI enthusiast with a background in Office Technology and Management from the Federal Polytechnic Offa. She went further to study an MSc in International Business at De Montfort University (DMU). With extensive work experience across administrative and business roles, she now focuses on exploring how artificial intelligence can transform work, innovation, and everyday life.