Dallas – NVIDIA announced a new Open Agent Safety Platform designed to give enterprises tighter control over artificial‑intelligence agents. The suite combines software policies with a hardware reference design, creating isolated environments that restrict what an AI can see or do.
How the platform works
The core of the offering is OpenShell, an open‑source software layer that runs on NVIDIA’s Vera CPUs. OpenShell lets operators define policies governing an agent’s access to files, tools, networks, processes and credentials. While the agent carries out tasks, the software monitors activity, enforces the rules and logs any violations.
OpenShell can be extended to run on third‑party platforms such as Arm and Intel, giving developers flexibility across data‑center hardware. A separate security component, NVIDIA Sentry, runs on BlueField‑4 data‑processing units. Sentry continuously watches agents, detects attempts to move beyond defined boundaries and can quarantine a rogue process within milliseconds.
Technical safeguards
The platform uses NVIDIA DOCA software to inspect every request and response, verify the agent’s identity and enforce access policies for data, tools, APIs and services. By separating security controls from the AI model itself, NVIDIA says the design makes it harder for malicious actors—or a misbehaving agent—to bypass protections.
More than 100 organizations have already joined the effort, including Anthropic, Cisco, CrowdStrike, Dell Technologies, JPMorgan Chase, Microsoft, Palantir, Palo Alto Networks, Salesforce, SAP, Scale AI and ServiceNow. Anthropic’s Claude Managed Agents integrate OpenShell and BlueField to add policy enforcement, while Salesforce has linked OpenShell with Slack so users can review and approve agent actions in real time.
Local relevance
The announcement follows NVIDIA’s 2025 plan to build AI supercomputers in Texas with local manufacturing partners. Those facilities are expected to boost high‑tech jobs and reinforce the state’s position as a hub for advanced computing.
Justin Boitano, NVIDIA’s vice president of enterprise AI, told CBS News that the platform could have stopped the recent Hugging Face breach, which was traced to an internal research model. “From what we know, this new security platform could have stopped the breach if it was being used in frontier labs for model evaluation early on,” Boitano said.
Industry impact
Robotics firms such as Figure, Gecko Robotics and Skild AI are adopting OpenShell to secure autonomous systems. Scale AI is incorporating the technology into its own agent infrastructure, and financial services firms including Citi and JPMorgan Chase are collaborating on open‑source safety tools.
OpenShell is available through NVIDIA’s developer portal and on GitHub under the Apache 2.0 license. The company says the platform is intended to foster cooperation on AI safety through shared tools, research and best practices, and it aligns with the Open Secure AI Alliance, a Linux Foundation‑governed initiative that now counts more than 120 members.
Looking ahead
CEO Jensen Huang emphasized that AI safety requires controls across the entire technology stack. By providing a modular, open‑source framework, NVIDIA hopes to give organizations—from startups to Fortune 500 firms—a practical way to keep powerful AI agents operating within safe, predictable limits.
Original reporting: The Dallas Express — read the source article.
