BLOG | Governance is not a feature. It's the whole job.

There was a moment, not long ago, when the plan seemed simple. Haul all your data to a data lake. Unleash an agentic harness on it. Let the AI figure things out.

"That lasted one hot minute."

Romit Ghose is not being unkind. He's being accurate. The assumption that enterprise AI could be governed from a single, centralized data architecture collapsed quickly when organizations started counting the actual number of agents operating across their estate — in-domain application vendors, on-premise installs, hyperscaler environments, third-party platforms. No single architecture covers all of it. And if your governance strategy assumes it does, you don't have a governance strategy.

The model is not the problem

A related assumption is proving equally stubborn: that trust in AI is fundamentally a model question. Get the right model, apply the right guardrails, and you're covered.

"That is a very 2023 thought," Ghose says.

The reality is considerably more complex. Demonstrably establishing enterprise-grade trust touches many layers — not just models, but context, how models are prompted, identity, access, and user management. The model is one component of a much larger stack. And the rest of that stack — the interfaces, the data flows, the agent interactions — carries its own risk surface that model-level governance simply doesn't reach.

Risk, security, and governance standards haven't changed just because the technology stack has. CIOs and CISOs are held to the same high standards. It's every piece of the stack delivering AI that matters, not just the model.

Governance as an aggregative act

This is where Ghose's framing gets most useful. In conversation with Guy Lupo, he drew a distinction that the industry needs to take seriously: governance is not a product feature. It's an aggregative act.

When agents are coming from multiple environments — on-premise, hyperscaler, in-domain — the only viable approach is a centralized one: discover all of them, hold them to the same standards, and apply a uniform layer of compliance, risk, security, and governance across all. That's the only way to hold every agentic experience and use case to account.

What that means in practice is an overarching toolset that sits above the complexity — not replacing it, but making it legible and governable. ServiceNow has built over 30 connectors to various agentic platforms. At the Innovate Americas hackathon, teams were able to aggregate agents from within ServiceNow and from AWS Bedrock, demonstrating how centralized governance can be applied across a heterogeneous agent environment.

The hackathon was significant beyond the demo. Forty teams integrated into a shared control tower, with assets and artifacts scanned well beyond the model level. It was, in Lupo's words, a proof point for what enterprise-grade trust actually looks like when you build it in code rather than describe it in a slide.

Building toward a standard

Standardization is beginning to emerge in the enterprise around what components need to be monitored, governed, and observed in order to certify an AI system as low risk. It's early — and Ghose is honest that no single standard has landed yet. But the direction is clear.

As part of that effort, ServiceNow is now contributing to TM Forum's Trustworthy AI and Data governance program, bringing recommended controls and risk frameworks into a shared trust profile that can be tested, validated, and adopted across the industry. It's the kind of practical, standards-based collaboration that moves the conversation from principle to proof.

ServiceNow has long been the CIO's cockpit for managing applications and assets across the enterprise. That same infrastructure, now extended with agentic capabilities, is what makes centralized governance viable in this new environment.

The AI control tower, in other words, isn't a new product category. It's a natural evolution of what good enterprise management has always required — visibility, accountability, and control at scale.

The agents are here. The governance has to catch up.

Watch the full conversation: Governance is not a feature. It's the whole job

Read Guy Lupo's blog: A hackathon that turned trustworthy AI ambition into working proof