NVIDIA CEO Jensen Huang and ServiceNow announced Project Arc, a new system that deploys autonomous AI agents capable of navigating desktop environments just like humans do. The agents are already available in early preview, powered by NVIDIA OpenShell for computing and runtime, while ServiceNow provides governance and workflow management. This partnership represents a significant milestone in enterprise automation, moving beyond simple task completion to handling complex, multi-system processes that require human-like reasoning and decision-making.

What Makes Project Arc Different From Other AI Automation Tools?

Project Arc stands out because it doesn't require companies to rebuild their workflows or integrate new systems. Instead, these desktop agents work within existing enterprise environments, mimicking how employees navigate applications, switch between tools, and complete tasks that span multiple platforms. This approach addresses a critical pain point: most AI agents today struggle with cross-ecosystem, end-to-end processes that require coordination across different software systems.

The partnership also introduced NOWAI-Bench, an open benchmarking standard for evaluating AI agents. This standardized approach allows enterprises to measure agent performance consistently and compare different solutions fairly, creating transparency in a market that has historically lacked clear performance metrics.

"ServiceNow is essentially the AI enterprise operating system," said Jensen Huang, CEO of NVIDIA.

Jensen Huang, CEO at NVIDIA

How Does Project Arc Fit Into ServiceNow's Broader AI Strategy?

ServiceNow has been building out what it calls an "Autonomous Workforce," a collection of specialized AI agents designed to handle specific business functions. Project Arc extends this vision by enabling these agents to operate at the desktop level, where much of actual enterprise work happens. ServiceNow's Control Tower, an AI governance platform, oversees all these agents, ensuring they operate within defined boundaries and don't cause unintended consequences.

The Control Tower itself represents a major evolution in AI management. Rather than simply monitoring what AI agents do, it actively intervenes when problems arise. During a Knowledge 2026 demonstration, ServiceNow showed how the Control Tower detected an AI agent that had been compromised through a prompt injection attack, automatically revoked its permissions, and shut it down before it could cause damage.

Steps to Understand How Enterprise AI Governance Works

• Discovery Phase: The Control Tower identifies all AI assets an enterprise uses, including models, agents, workflows, cloud platforms like AWS and Azure, and enterprise applications such as SAP and Oracle, even those outside of ServiceNow's direct control.
• Observation and Monitoring: Real-time tracking of agent behavior with immediate alerts when anomalies occur, supported by deeper visibility into how agents reason and make decisions through tools like Traceloop.
• Governance and Risk Management: AI-driven risk assessments across agents, models, prompts, and datasets, with automated remediation when agents operate outside established risk frameworks aligned with EU AI Act and NIST standards.
• Security and Access Control: Mapping of who or what has access to systems, with automatic shutdown if agents attempt to act outside their assigned permissions.
• Measurement and ROI Tracking: Token-level and workflow-level usage tracking linked to business outcomes, allowing companies to scale successful AI implementations and prevent runaway spending.

Why Should Enterprises Care About Desktop Automation Now?

The enterprise software market has spent years chasing the promise of "autonomous" AI agents, but 2025 revealed a hard truth: most agents can't handle real-world complexity without human oversight. The problem isn't that the AI doesn't work; it's that enterprise data is messy, governance frameworks are unclear, and processes span multiple incompatible systems.

Project Arc addresses this by operating at the desktop level, where human workers already know how to navigate complexity. Rather than forcing enterprises to restructure their workflows around AI capabilities, the agents learn to work within existing processes. This pragmatic approach acknowledges that enterprise transformation takes time, and automation can begin immediately without waiting for perfect data or complete system integration.

ServiceNow CEO Bill McDermott framed this shift as moving from "chaos to control," emphasizing that governance and visibility matter as much as raw AI capability. For enterprises worried about losing control of their systems or facing security risks from autonomous agents, the Control Tower's ability to actively intervene provides a safety mechanism that didn't exist before.

The announcement also signals confidence from NVIDIA, one of the world's most influential AI infrastructure companies, in ServiceNow's approach to enterprise AI. Huang's public endorsement carries weight in a market where enterprises are still deciding which platforms to bet on for their AI transformation.