When Jamie Robbins and his colleagues at NTT Data began mapping the AI running across the company, they found five separate productivity agents answering the same employee questions, each commissioned by a different part of the business and each carrying its own costs.
The years of acquisitions have left NTT Data with around 2,500 applications and nine HR systems; every business it bought arrived with its own tools, and employees then added AI of their own choosing. “AI is growing faster than anyone I think we predicted, and it’s growing faster than we can see it,” said Robbins, Head of the ServiceNow Practice at NTT Data.
Other ServiceNow customers have met the same problem at different scales. LTM has deployed 1,500 AI agents and given each one an employee ID and a manager, borrowing the accountability structure it already uses for its people. Chandan Pani, Chief Information Security Officer at LTM, said ServiceNow’s Super Intelligence Control Tower now acts as “a single system of record for these AI agents, giving us centralised visibility into inventory, risk, and compliance”.
At PepsiCo, which is expanding AI across operations from supply chain to service delivery, the pressure comes from sheer reach. Maximiliano Garnica Pineda, AI and Innovation Lead at PepsiCo, pointed to a business spanning more than 200 countries and territories and over 350,000 employees. “We want agentic AI in every corner of that business. A governance foundation is what makes scaling to that level possible,” he said.
ServiceNow says a kill switch means little until an agent’s limits are written down
Those customers were cited as ServiceNow announced AI Passport on 15 October, the latest addition to the Super Intelligence Control Tower it uses to discover, monitor and govern AI across an enterprise. Speaking at a briefing ahead of the launch, Nenshad Bardoliwalla, Senior Vice President of AI Products at ServiceNow, said the popular image of a single red button misread how control works in practice. “Real governance is graduated. It’s more like a dimmer switch. You monitor, you pause, you restrict, and eventually, if necessary, you deactivate. But you can’t enforce any of those controls without knowing what the agent is authorised to do in the first place,” he said.
AI Passport supplies that baseline. Each agent receives a cryptographically signed credential issued by the Super Intelligence Control Tower, recording which systems it can reach, which actions it may take and who is accountable for it, drawn from the configuration management database, compliance policies and workflow approvals the enterprise already keeps in ServiceNow. The specification is open and covers third-party agents as well as ServiceNow’s own, which Bardoliwalla said gave it the potential to become an industry standard. Jon Sigler, Executive Vice President and General Manager, AI Platform at ServiceNow, said that as agents spread, visibility on its own was no longer sufficient: “Enterprises need confidence that every AI agent is operating within its authorised scope.”
NVIDIA will check the passport inside the infrastructure where agents run
NVIDIA is the design and launch partner, building enforcement into its Open Agent Safety Platform, which includes the OpenShell secure runtime and the NVIDIA Sentry security framework. Itay Ozery, Director of Product Marketing for Networking and Cybersecurity at NVIDIA, told the briefing that enterprises had seen what agents could deliver and now wanted them at far greater scale, which meant trusting them with more. “As agents take on more work, they also need more authority. They’re accessing enterprise data, using tools, delegating tasks to other agents, and really taking actions that affect the business,” he said.
The identity systems, access controls and approval chains most companies already run were designed around people and applications with fairly predictable behaviour, Ozery explained, whereas an agent chooses tools mid-task and passes work to other agents, so its permissions have to travel with the work without widening along the way. Enforcing those limits at the infrastructure layer also gives an independent safeguard if an agent escapes its software boundary or its host is compromised. “The hardest part is really making enterprise policy meaningful at the point of execution,” he said.
ServiceNow demonstrated the mechanism with an off-the-shelf customer service agent capable of five tasks, which it restricted to processing refunds with a $25 ceiling. When two requests arrived, the $18 refund went through automatically and the $40 refund was stopped at runtime and sent to a human. Bardoliwalla said most customers would begin with human approval on decisions like these and let agents act alone only once they had proved reliable. AI Passport is expected to be available in Q4 2026.
EmployeeWorks hands staff their own agents on the company’s terms
The second part of the announcement extends EmployeeWorks, the employee assistant ServiceNow launched in February, built on the Moveworks technology that Bhavin Shah, Senior Vice President of the Employee Experience Group at ServiceNow, helped create before it joined the company. Shah said most AI harnesses on the market chased developer productivity, and ServiceNow had built one around employees’ roles, departments and projects, designed to stop AI acting “in what I call YOLO style” and hold it to “the rules that you’ve established, the business processes”.
The update adds productivity assistants tailored to specific roles, such as a recruiting assistant for hiring managers; productivity routines, which let employees build no-code agents for repeatable tasks; a Conversations API that carries EmployeeWorks into internal apps, kiosks and third-party AI agents; and Toolsmith, which lets developers describe the connectors they need in plain language and then builds, tests and debugs them using connections the organisation has already approved.
Shah called Toolsmith “probably one of the best things my team and I have built over the last 10 years”, and ServiceNow claims it cuts build time from weeks or months to minutes, though it has not published customer data to support the figure.
NTT Data is putting EmployeeWorks over the systems it has yet to consolidate, so staff can log an AI use case, raise an IT request or book leave in one place. Today, a manager whose team sits on different HR platforms may have to open three to see who is on leave. “By the time we get everyone on the same system, it could be another 18 months potentially to create that visibility that we can create maybe tomorrow,” Robbins said.
Toolsmith is the addition he is most interested in, because it changes the size of team his transformation work requires. “I can have a single developer build me these capabilities and these experiences very quickly and test them, troubleshoot them,” he said.
The customers’ next test is whether governance can cut waste as well as risk
LTM expects the Super Intelligence Control Tower to improve operational efficiency significantly by streamlining how it manages AI assets, though it gave no figures. Pani said the capabilities “will be foundational to scaling AI responsibly across LTM, enabling stronger governance, enhanced risk oversight, and greater business value from our AI investments”.
Robbins wants the platform to go further at NTT Data and prevent the kind of duplication that produced those five overlapping agents, checking each new AI request against what the company already runs and steering employees to an existing tool before anyone builds or buys another. “We’re actually starting to say, well, this already exists, this is where you can use this, so we actually start to direct them to value quicker as well,” he said.
