From Local Agents to Enterprise AI Governance
Enterprise AI governance is evolving from a focus on individual tools to holistic oversight of autonomous systems. NVIDIA and ServiceNow are pushing this transition by extending AI governance from desktop-based agents to the underlying data centre infrastructure that powers them. The partnership centres on governed AI deployment, where every action taken by an AI agent—whether generating code, manipulating data, or calling APIs—occurs under auditable, policy-driven control. This marks a shift from ad hoc experimentation with AI on personal machines to centralized AI agents security enforced by IT. By aligning AI controls with existing operational platforms and configuration management databases, enterprises can embed AI agents into core workflows without surrendering visibility or compliance. The result is an emerging model of data centre AI oversight that treats AI workloads like first-class, governable infrastructure components rather than opaque, standalone tools.
Project Arc: Autonomous Desktop Agents in a Controlled Shell
At the desktop layer, ServiceNow’s Project Arc exemplifies how governed AI agents can operate autonomously without compromising control. Arc is designed to write code, execute complex, multi-step tasks, and adapt to changing conditions across enterprise applications—all without depending on rigid, pre-built workflows. Its autonomy is bounded by a secure runtime built on NVIDIA OpenShell, which sandboxes the agent and enforces policy-based constraints on what it can access or execute. ServiceNow’s AI Control Tower provides continuous monitoring of the agent’s behaviour, logging files touched, commands run, and external APIs invoked. Integrated with ServiceNow’s Action Fabric and CMDB, Arc can leverage operational history and system data while remaining fully traceable. For IT teams, this offers a blueprint for AI agents security at the endpoint: powerful, self-directed agents that still operate inside a governed and auditable environment suitable for regulated and mission-critical use cases.
Extending Governance into Data Centre AI Infrastructure
The most significant change for enterprise AI governance comes from extending oversight into the data centre itself. By integrating ServiceNow’s AI Control Tower with NVIDIA’s Enterprise AI Factory validated design, governance now spans not just agents but the AI infrastructure that runs them. This integration enables model discovery, centralized inventory of AI workloads, and detailed observability of how models behave in production environments. Compliance monitoring is built in, supported by regulatory content packs and frameworks that map cloud access and track runtime costs and productivity gains. For enterprises, this creates data centre AI oversight that aligns security, compliance, and resource management in a single control plane. Long-running, autonomous agents can be deployed at scale with confidence, as IT teams gain the ability to detect policy violations, trigger remediation workflows, and ensure that AI operations conform to internal and external regulatory standards.
Centralized Control for Distributed AI Workloads
As AI workloads spread across desktops, on-premises clusters, and cloud environments, enterprise IT faces a distributed management problem. The NVIDIA–ServiceNow collaboration addresses this by turning AI Control Tower into a command centre for governed AI deployment across heterogeneous infrastructure. IT teams can monitor AI agents running on individual workstations alongside model workloads executing in data centres, using consistent policies and dashboards. Cloud access mapping clarifies which AI services are used where, while observability tools surface performance and reliability issues before they impact users. This centralization reduces the risk of shadow AI deployments and improves audit readiness by consolidating logs and compliance evidence. In practice, enterprises gain a unified view of AI risk, usage, and value. AI agents security and governance become woven into existing IT service management processes, allowing organizations to scale AI initiatives without losing operational discipline or regulatory alignment.
Open Benchmarking and the Future of Accountable AI Agents
Governance is only as strong as the metrics used to evaluate AI behaviour. To support accountable AI agents, NVIDIA and ServiceNow are advancing NOWAI-Bench, an open-source benchmarking suite tailored to enterprise scenarios. EnterpriseOps-Gym focuses on multi-step workflows in IT service management, customer service, and HR, while EVA-Bench evaluates enterprise voice agents. Integrated into NVIDIA’s NeMo Gym platform, these tools allow organizations to stress-test AI agents under realistic, complex conditions before widespread rollout. This benchmarking complements data centre AI oversight by providing standardized performance and reliability baselines that can be tied to governance policies. Over time, such open benchmarks could help establish industry norms for safety, robustness, and compliance in enterprise AI. Together with centralized control and secure runtimes, they point toward a future where autonomous agents are not only powerful and ubiquitous but also measurable, transparent, and accountable within enterprise governance frameworks.