The Board Changed the Question.
Can CISOs Answer It?
Security leaders are moving beyond traditional strategies to govern autonomous systems at machine speed.
Not long ago, the most common question boards of directors asked chief information security officers was simple and direct: Are we secure?
Today, that conversation has evolved into something much harder, more specific and far more urgent. Boards no longer just want to know if the perimeter is locked down. They need an answer to an almost unknowable question: Can we trust the AI technology that’s running our business? ServiceNow’s study on the state of agentic security shows why that question matters. Fifty-six percent of leaders rate their agentic security as established, yet less than half (46%) have a security framework purpose-built for agentic AI.
That single question is reshaping what it means to lead security in the modern enterprise. The hurdle isn’t that security teams are unprepared. It’s a structural mismatch.
Most traditional security processes were built for a human cadence, including periodic reviews, approvals and a person in the loop to make any consequential decision. But AI agents don’t run on a human cadence. They operate continuously, at machine speed, across complex webs of systems that no single department fully owns. The CISO’s job has become exponentially harder.
AI Compounds the Attack Surface
Agent scale compounds. Governance is still added one review at a time. The gap creates exponential exposure
run AI agents in, or interacting with, operational technology or physical security environments
of organizations operate more than 1,000 AI agents
29%
discovered AI agents already running in production that were never registered or known
20%
have a single named accountable owner for agentic AI governance
Source: “Deployed but Ungoverned: The State of Agentic Security,” ServiceNow, October 2026.
Rebuilding for
Machine Speed
Ben de Bont, the CISO of ServiceNow, points out the fundamental hurdle: AI agents operate in both deterministic and nondeterministic contexts, and you cannot design rigid rules for scenarios you cannot predict. “The governance cadence of the past is built on the assumption that you can design for everything up front, and then you can have it easily audited by a human,” de Bont says. “And when you have thousands of autonomous agents running at once, a governance model or architecture built around periodic approval is outmatched.”
To scale safely, de Bont champions a shift from a “human in the loop” model, where a person is expected to weigh in on all decisions, to a “human on the loop” approach, where a person sets the agent’s intent and reviews outcomes, directly intervening only when a critical issue arises. “Only one of those scales when threats are moving at machine speed,” he says.
– Ben de Bont, CISO, ServiceNow
“The governance cadence of the past is built on the assumption that you can design for everything up front, and then you can have it easily audited by a human. And when you have thousands of autonomous agents running at once, a governance model or architecture built around periodic approval is outmatched.”
ServiceNow has built this philosophy into its operating model through what it calls Shift Zero, a framework for governed, autonomous security: the ability to deploy AI faster than threats scale, and still answer with proof what every system in an environment is doing, why it’s performing those actions and who is accountable. Shift Zero’s core principle is prevention before manifestation, architected for greater cyber resilience. ServiceNow’s Agent Incident Response solution is one expression of that approach, automating routine triage, investigation and containment where playbooks exist. The company built the platform for its own use before offering it to any customers, learning through experience that governance is what makes autonomous action trustworthy.
The sheer volume of nonhuman entities that CISOs must keep an eye on is staggering. According to ServiceNow’s study, 35% of organizations already operate more than 1,000 AI agents. Yet only 57% of organizations report a complete, accurate inventory of their agents. And roughly a third of organizations have discovered AI agents running in production that were not previously known or registered. The stakes are high. Ninety percent of organizations reported an agentic-security incident in the past year, and only 36% of those organizations contained the incident within 24 hours. “The board should be aware of all of the challenges around nonhuman identities and ensure there is accountability,” says Maxine Holt, vice president of enterprise and channel research at Omdia. “I think the risk of not doing anything, not adopting AI agents, is often seen by organizations as a bigger risk than the security aspect.”
Autonomous
Workforce Booms
A Defensible Security Posture Delivers Greater Results
Most organizations sit short of a defensible posture. But those that have one are seeing clear security benefits
All others
Defensible postures
Automated real-time anomaly detection
Contain incidents in under 24 hours
Mandatory automated deployment gates
Real-time visibility into 90%+ of agents
0%
10%
20%
30%
40%
50%
60%
70%
49%
65%
33%
55%*
29%
49%
17%
31%
Source: “Deployed but Ungoverned: The State of Agentic Security,” ServiceNow, October 2026.
McHenry notes that AI agents deployed to defend a company hold a distinct advantage: They know their “house.” While an AI attacker might swarm quickly with breadth, defenders understand the exact “rooms” to protect.
And, ironically, he says the rise of agentic threats has created a new level of respect for CISOs.
“The CISO role can be terrifying,” McHenry says. “You own risks that you’re not responsible for. But over the past six months, it’s been refreshing. It’s generally understood across all of humanity that AI agents and AI adoption are risky—and it is a societal turning point for us, and an opportunity for every CISO to step up and show the industry the safe way to adopt new technology.”
Competitive
Advantage
While governance provides the foundation, progressive organizations are learning to harness AI transformation as a strategic edge. At Robinhood, the business has evolved rapidly to become an AI-native company, where almost 100% of employees are using AI tooling and 50% of the code base is AI-generated.
The company has also embraced agentic security, says Evan McHenry, Robinhood’s vice president and CISO. “At this point, we have well over 60 different AI agents running different parts of security and enterprise technology at Robinhood,” he says. “This is fully autonomous and backed by our policies.”
Achieving the right balance of speed and safety requires rigorous alignment on security, compliance and business goals before pilots even launch. Mahmood Khan, senior vice president and global chief security officer at CNA, emphasizes that enterprises must focus on practical execution.
“The real test for AI is not whether we can make it work,” he says. “It’s whether we can make it work in a way that is trusted, scalable and produces a measurable business outcome. That’s where I believe enterprises need to focus now.”
To maintain this trust, security must carefully regulate what agents can access, what they can do independently and when human intervention is necessary. Continuous review is nonnegotiable.
Khan warns security teams must watch closely for specific red flags: “An agent accessing information it does not need, interacting with systems it was not supposed to access, suddenly operating at a much higher volume or bypassing a normal business process.”
Ultimately, the evolving mandate for CISOs is clear. By treating AI governance as core infrastructure rather than an afterthought, organizations can move past the limitations of periodic reviews and answer the board’s toughest question with clear proof: Yes, they can trust the autonomous systems running their business.
Scaling
Trust
Read ServiceNow‘s report on the state of agentic security.
LEARN MORE
This content was created by Custom Content from WSJ, a unit of The Wall Street Journal Advertising Department.
90%
experienced an agentic-security incident in the past year
only 36% contained the agentic-security incident within 24 hours
35%
82%
Reported Security Measures Achieved
*Among organizations that reported an incident.