ReadingResearchRadarInvestment framework
Sign in / Sign up
← Back to reading
NewsMaterial2 min readDisconfirmAI Editorial

ServiceNow and NVIDIA Announce at GTC Berlin: Arc AI Agent Generally Available, Governance Policies Extended to BlueField Hardware Layer

ServiceNow launches its Arc personal AI agent and partners with NVIDIA to extend AI Passport permission validation from software runtime to BlueField chips, enabling sub-second hardware-level enforcement.

Verified 2026-10-22 02:45 GMT+8

At the NVIDIA GTC conference in Berlin on October 21, 2026, U.S. enterprise software company ServiceNow (NYSE: NOW) announced the general availability of "Arc," a personal, long-running AI agent designed for employees. As an expansion of its partnership with NVIDIA, ServiceNow introduced a new AI governance architecture that extends permission controls from traditional software layers down to the hardware level, aiming to address the trust and control challenges enterprises face when deploying autonomous AI agents at scale.

According to the official press release, Arc is an AI agent capable of handling everyday tasks, planning work, and executing actions with limited human involvement. It can run on an employee's desktop or within a secure sandboxed runtime provided by NVIDIA OpenShell. The core of this update lies in the deep integration of the "AI Passport" solution with NVIDIA hardware. AI Passport is a cryptographically signed credential issued by the ServiceNow Super Intelligence Control Tower, defining the scope of tools, APIs, and data each AI agent is authorized to access.

Previously, policy enforcement for AI agents relied primarily on software-layer monitoring and interception. In the new architecture, both the NVIDIA OpenShell runtime and the Sentry security framework on NVIDIA BlueField Data Processing Units (DPUs) directly read the status of the AI Passport. This means that when the Super Intelligence Control Tower revokes an AI agent's permissions, NVIDIA BlueField can enforce this decision at "machine speed" at the hardware layer, blocking the agent's network traffic or operations without requiring host infrastructure restarts. Officials state that this cross-layer revocation propagation takes approximately one second.

"As AI agents take on more autonomous tasks, governance and security have to extend to everything they touch," said Jon Sigler, Executive Vice President and General Manager of AI Platform at ServiceNow. Justin Boitano, Vice President of Enterprise at NVIDIA, noted that through OpenShell, the governance defined by ServiceNow can be extended into the runtime layer where agents actually take action.

Sign in / Sign up
ReadingResearchRadarInvestment framework

Beyond hardware-level enforcement, Arc also demonstrates capabilities for local, lower-cost inference using NVIDIA RTX Spark, allowing everyday tasks to be handled by local models while reserving frontier models for complex demands, thereby reducing inference costs and enhancing data privacy. Arc is currently included in ServiceNow EmployeeWorks and available at no additional cost to customers with Foundation, Advanced, and Prime SKUs.

Sources

  • newsroom.servicenow.com↗
  • markets.ft.com↗
Continue researching →