IBM and NVIDIA Introduce Advanced Identity and Security Frameworks for Enterprise AI Agents

Image: IBM · Source

IBM unveils AI agent identity and control features with HashiCorp Vault integration, while NVIDIA launches OpenShell Reference Architecture to ensure secure, isolated, and accountable AI agent operations across enterprise infrastructures.

At TechXchange 2026, IBM announced significant advancements enhancing the governance, identity, and security of AI agents deployed across enterprise systems. With enterprises anticipating over 150,000 AI agents in use by 2028, IBM integrates agent discovery, identity verification, access control, and performance evaluation to offer a comprehensive framework for managing AI agent authority and accountability. Central to this is the extension of HashiCorp Vault capabilities to AI agents, providing securely managed, short-lived credentials that enforce least-privilege access and enable real-time revocation to minimize security risks. IBM’s Agent Identity establishes formal records linking agents to their owners and authorized roles, ensuring clear governance. Complementing these controls, IBM’s watsonx Orchestrate enables discovery and sandbox containment of agents spanning multiple platforms like Amazon Bedrock, Microsoft Foundry, and Google Gemini, while offering evaluation tools for quality, compliance, and cost monitoring. Furthermore, IBM integrates these solutions with business systems via its Integration Intelligence Hub, furnishing agents with context-aware operational knowledge and compliance checks. In parallel, NVIDIA introduced the OpenShell Reference Architecture designed to provide enterprise-grade governance, isolation, and security for AI agents operating both in data centers and at the edge. Leveraging NVIDIA’s OpenShell secure runtime software alongside specialized hardware components such as NVIDIA Vera Rubin infrastructure for high-throughput inference, Vera CPUs for agent harnesses, and BlueField-4 DPUs running the Sentry security framework, this architecture offers continuous behavioral monitoring and in-silicon enforcement to prevent unauthorized agent actions. OpenShell supports a range of enterprise operating systems and cloud environments, enabling the deployment of AI agents from diverse model providers under tight security protocols. NVIDIA’s collaboration with major industry players spans cybersecurity vendors, banks, software companies, and cloud providers, all adopting OpenShell to enhance AI agent security and governance. Together, IBM’s identity-focused platform and NVIDIA’s secure runtime architecture mark a coordinated advance towards safely scaling AI agents in complex enterprise ecosystems, addressing emerging risks related to credential management, unauthorized access, and operational control. Both sets of innovations are generally available or in various stages of public preview, underlining the industry’s commitment to integrating AI agent capabilities with robust identity and security frameworks.