ServiceNow extends Nvidia governance

Enterprise AI is moving past chat windows. The real fight now is over autonomous agents that can click, type, call APIs, and trigger work on their own. That is useful — but it is also where companies get nervous, because the same agent that saves time can also touch files, credentials, and systems it should not. ServiceNow’s latest move with Nvidia is basically an attempt to make that risk manageable, not just on a worker’s desktop, but all the way down in the data-center layer. (newsroom.servicenow.com) ### What actually launched? On May 5 at ServiceNow’s Knowledge 2026 event in Las Vegas, the two companies announced an expanded partnership built around three pieces: Project Arc, Nvidia OpenShell, and ServiceNow AI Control Tower. Project Arc is the new thing users will notice — an autonomo(newsroom.servicenow.com)rsight layer that governs those actions. (newsroom.servicenow.com) ### Why does Project Arc matter? Most enterprise AI tools still wait for a prompt. Project Arc is meant to act. ServiceNow says it can live on a desktop and autonomously complete complex work, which is a different category from a chatbot or copilot. Once software starts acting inside real workflows, governance stops being a nice add-on and becomes the product’s safety system. (newsroom.servicenow.com) ### What is OpenShell doing here? OpenShell is Nvidia’s open-source runtime for autonomous agents. The simple way to think about it is as a browser-style sandbox for AI workers — every session is isolated, permissions are checked before execution, and policies can restrict file access, netw(newsroom.servicenow.com)y incident. (github.com) ### What does ServiceNow add on top? ServiceNow is not just supplying another agent. It is trying to be the control plane. AI Control Tower governs which agents are allowed to run, what policies apply, and what gets logged for audit and compliance. In this partnership, that layer now extends beyond the desktop-agent use case and into Nvidia’s Enterprise AI Factory validated design, which means governance is being pushed (github.com)kloads run. (newsroom.servicenow.com) ### Why is the data-center angle the bigger story? Because that is the shift from “AI app” to “AI system.” Governing one agent on one machine is useful. Governing fleets of agents and model workloads inside enterprise infrastructure is where budgets get much larger and where platform lock-i(newsroom.servicenow.com)the estate. (newsroom.servicenow.com) ### Is this only about security? Not really. Security is the first reason buyers care, but observability is the second. If an agent accesses the wrong file, runs the wrong command, or hits the wrong API, companies need an audit trail. That is why this announcement also included an open benchmarking effort for AI agents — the companies said NOW AI Bench is meant to help standardize how enterprise agents are evaluated, with Nvidia already using it on Nemotron models. (businesswire.com) ### Where does this fit in ServiceNow’s bigger push? It fits neatly into ServiceNow’s argument that AI intelligence is becoming cheap, while governance is becoming scarce and valuable. At Knowledge 2026, the company tied AI Control Tower to a broader strategy around autonomous work and said it is targeting more t(businesswire.com)tic AI from chaos into something a CIO will sign off on. (msn.com) ### Bottom line? This is not just a desktop-agent launch. It is a claim that enterprise AI needs a governed runtime plus a governed control layer, and that those controls have to reach from the employee screen to the model factory in the data center. If that idea sticks, the winners will not just be model builders — they will be the companies that decide what autonomous agents are allowed to do. (newsroom.servicenow.com)