Bill McDermott, ServiceNow's chief executive, said 94% of companies have zero to very little visibility into the AI running inside their own business.
Warnings about rogue AI have been piling up from researchers and ex-employees of the big labs. McDermott's pitch is that the fix isn't slowing AI down — it's an "AI control tower" that can see every model in use and shut one off if it goes rogue.
"AI can do a lot of things for you. It can also do a lot of things to you."
McDermott has built ServiceNow into what he calls the fastest-growing enterprise software company in the world, and now, on the back of a run of acquisitions, the fastest-growing enterprise security company too.
I listened to the full segment so you can skip it.
Here are the 4 calls that matter.
👤 Guest: Bill McDermott, chairman and chief executive of ServiceNow
🎙️ Hosts: Sarah and Karl, co-anchors of Squawk on the Street on CNBC
📰 Published: 10 September 2026 on CNBC's Squawk on the Street
🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 6 min
Key Takeaways
94% of companies have zero to very little visibility into the AI running inside their business Most can't even tell personal AI accounts apart from corporate ones
ServiceNow's AI control tower is built to see, govern and — if needed — kill rogue AI agents It applies more than 400 global AI regulations as rules and rails for running the company
The real AI value is moving up the stack, from infrastructure into execution He put the total cost of ownership of software built from scratch on large language models at up to ten times a proven platform's
The Armis, Vasa and ServiceNow combination gives one control tower over identity, assets and workflow On one public-sector client, it found 95% of assets were unrecognized in its own graph
1. The Visibility Gap
McDermott opened with the risk before the pitch. "AI can do a lot of things for you. It can also do a lot of things to you. And that's why ServiceNow is the lead enterprise software company in moving from workflow to cybersecurity."
He put a number on how blind most companies are. "94% of the companies out there have zero to very little AI visibility, and most of them cannot distinguish between personal AI accounts or corporate instances. That's real bad."
His case for the AI control tower rests on three steps. "You have to see it first," then govern it against "the 400 plus global AI regulations that are already there," then hold a kill switch for "rogue agents that get out of control" The economics piece is separate again: "Track the token consumption, enforce the budgets, and measure AI ROI"
2. How the Kill Switch Works
Asked to explain the kill switch, McDermott tied it to two decades of ServiceNow's own system-of-record data. "A lot of companies will call us the ERP of IT. So we've had the system of record for more than two decades of all the people, places and things that go across an enterprise."
He named the specific failure mode he's selling against: vibe slop, code built from scratch on large language models. He said it carries "architectural weaknesses that are turning today's productivity gains into technical debt and risk" — with a total cost of ownership he put at up to ten times a proven platform's
The system is meant to recognize misbehaving agents the way it would a misbehaving employee, and shut them down first. "It will recognize an agent when an agent, just like a human, is doing something that they shouldn't do... And then a human has to come in if they want to turn them back on. So we prefer to shut things down before mistakes are made."
3. The Deals Behind It
McDermott connected the strategy to ServiceNow's Armis and Vasa deals, which drew skepticism at the time. "We were doing it because we wanted to have the one and only integrated cyber security architecture that connects identity, assets, AI, and workflows because we knew the market needed it." He described the split: "Vasa is the identity graph. Armis is the asset graph, and ServiceNow is the workflow graph."
His example of what the combination catches: a public-sector client that had no idea what it owned. "We can go in, for example, to a public service entity and identify the fact that they have 95% of their assets unrecognized in that graph."
He called the combination unique in the market. "Only company in the world that does it. Think about integrated risk management, security operations, identity management, and asset management, it, IoT, all things that move in an enterprise. We got it under control all on one platform."
4. Where the Money Moves Next
McDermott said the AI-spending story is shifting from infrastructure to execution. "I see the AI value moving up the stack from infrastructure, which has been heavily invested in... these labs, they're extraordinary in their intelligence, but they're also extraordinary in their cost. So now we're moving into the execution phase."
His formula for where ServiceNow captures that value: "We add proprietary enterprise context. AI doesn't work well without context... And then we apply workflows and governance and turn the intelligence into action. And that's where the money is."
Action Fabric is his name for the product doing that work. "Action Fabric is the gateway for the enterprise. So you have one secure entry point for any ServiceNow or third party agent to execute across the enterprise... using existing workflows and governance with no custom integrations."
Bonus Insights
McDermott name-checked the specific report that had prompted the host's question, a warning from an unnamed former Anthropic researcher who resigned sounding an alarm about out-of-control AI, without disputing its substance directly.
McDermott's bottom line is that the AI money is moving from the labs that build the models to the platforms that can see, govern and safely act on them inside a company — and that ServiceNow's two decades of enterprise data is what lets it make that shift first.
Products, Companies & Tools Mentioned
ServiceNow (McDermott's company, and the platform behind the AI control tower, Action Fabric and the autonomous IT service management kill switch he describes)
Armis (ServiceNow's asset-graph acquisition, one leg of the identity-asset-workflow control tower)
Anthropic (Where an unnamed former researcher's public warning about AI danger is what prompted the host's question)
If this was worth your time, send it to someone who follows the name.
Get the latest market chatter as it happens: