Agentic AI governance is moving beyond individual systems as ServiceNow and NVIDIA broaden their collaboration to connect desktop-level automation with data center infrastructure. The update introduces Project Arc, an enterprise-focused desktop agent designed to independently handle complex, multi-step tasks. Operating directly on employee devices, the agent can generate code, execute workflows, and adjust its actions in real time without relying on predefined processes.

All operations performed by Project Arc are executed within NVIDIA OpenShell, a controlled runtime environment that enforces policy-based safeguards. Oversight is handled by ServiceNow AI Control Tower, which tracks every interaction, including accessed files, executed commands, and API usage. This combination ensures that enterprise teams retain full visibility and control, enabling auditability and compliance across autonomous actions performed on desktops.

The partnership also extends governance into large-scale AI infrastructure through the integration of AI Control Tower into the NVIDIA Enterprise AI Factory validated design. This move brings centralized oversight to environments running extensive model workloads, allowing organizations to manage risk, compliance, and performance throughout the lifecycle of AI systems. Capabilities include monitoring model behavior, maintaining inventory, and applying corrective measures when needed.

Additional tools introduced for AI factory users include regulatory content packs, access mapping across major cloud providers, and frameworks to measure operational costs and value generation. These features aim to provide continuous insight into both the financial and operational performance of AI deployments, supporting decision-making at scale.

To support industry-wide evaluation, the companies are also releasing an open benchmarking suite. The initiative includes EnterpriseOps-Gym, which tests multi-step agent performance across enterprise workflows, and EVA-Bench, focused on voice-based AI systems. Both frameworks are available as open-source tools and are being integrated into NVIDIA’s NeMo Gym platform to standardize automated model assessment across use cases.

📊 What This Means (Our Analysis)

This expansion ties together two layers of enterprise AI that often operate separately: user-facing automation and backend infrastructure. By linking desktop agents with centralized governance, the partnership creates a consistent control framework across how work is executed and how AI systems are managed behind the scenes. That alignment addresses a core enterprise need—visibility across increasingly autonomous systems.

Equally important is the introduction of open benchmarks, which signals a shift toward measurable performance standards in enterprise AI. By making evaluation frameworks publicly available, the collaboration moves beyond deployment and into accountability, giving organizations clearer ways to assess outcomes and compare systems within a structured environment.

📌 Our Take: As enterprise AI grows more autonomous, control and measurement are becoming just as critical as capability.