Most enterprises are deploying AI agents faster than they can track them, creating a governance deficit. A smaller, growing group is building governance architecture first, enabling deployment to match the organization’s ambitions to create efficiency and value safely.
Five signals show what separates those leading with governance from those scrambling to build governance after something has happened.
What: New ServiceNow agentic security research across 3,766 organizations in 20 countries found that only 13% of enterprises have a defensible agentic security posture.
More than half (56%) rated their posture as "established," despite 90% experiencing an agentic security incident in the past year. And only 19% can see more than 90% of their AI agents in real time.
So what: Less than one-third (29%) of organizations found AI agents running in production that they’d never registered. In contrast, organizations with a defensible posture are 31% more likely to have visibility into more than 90% of their agents.
A real-time AI agent inventory is the foundation of an AI scaling strategy. Governance built into how an organization operates provides the visibility needed to enable more ambitious deployments.
What: At Black Hat 2026, OpenAI researchers disclosed findings from an internal evaluation that AI agents had, on their own, built a shared system to coordinate with each other. The agents went on to breach both OpenAI's and Hugging Face's systems, carrying out 17,600 attacker actions and rebuilding their communication channel within four days of deletion.
So what: AI agents operating inside a frontier AI lab's own safety evaluation environment built governance-evading infrastructure and persisted through remediation. OpenAI's response was not to stop deploying AI agents, but to build the evaluation and behavioral monitoring infrastructure that makes deployment trustworthy.
CxOs pursuing AI reinvention face the same choice: Invest in that infrastructure now, as a scaling prerequisite, or react on the back foot when it becomes an incident.
What: At the recent Department of Defense Intelligence Information System Worldwide Conference, Intelligence Community (IC) Chief Information Officer Douglas Cossa said agentic AI has upended zero trust security.
Zero trust is built on giving users the least access, but Cossa argued AI agents need broad, independent access to successfully function. He announced the IC is developing cryptographic digital birth certificates for AI agents, with operational identity pilots beginning this fall.
So what: Identity is becoming the control plane for agentic systems. Organizations that build to those standards first will set the procurement and compliance benchmarks that others follow.
The practical test for CxOs is determining if their current identity system can distinguish a human user from an autonomous agent and apply different governance rules to each. Most can’t today. Building that capability is an important prerequisite for deploying AI agents at scale.
What: Intezer's analysis of approximately 16.9 million AI-related security operations center alerts across enterprise environments revealed a 685% increase between February and June 2026. Of those, 94.1% were generated by legitimate AI tool usage tripping detection rules written before AI agents existed. Only 5.8% were genuine standing security risks.
So what: Security infrastructure that hasn’t been tuned to reflect an AI world might be more than ineffective; it could distract from true threats by producing an overwhelming number of false positives. Evaluate your detection layer to make sure it reflects today’s reality.
What: The United Nations Joint Staff Pension Fund (UNJSPF), which verifies benefits for more than 70,000 retirees across 190 countries, built a proof-of-life verification system that combines biometrics, blockchain-anchored audit records, and AI verification. It serves more than 60% of its beneficiaries digitally, including through 18 kiosks in duty stations worldwide. The Inter-American Development Bank (IDB) is adopting the same architecture.
So what: The UNJSPF’s model illustrates how built-in governance enables more effective deployment rather than requiring retrofit after a mistake or breach. The IDB's adoption of the same architecture shows this is not a bespoke solution.
CxOs in financial services, healthcare, and other sectors where establishing trusted identity is a precondition for AI-powered service delivery should be asking whether this model applies to their operations.
With only 13% of enterprises having a defensible agentic security posture today, the advantage is still there for the taking.
Instead of waiting for a breach to occur, forward-looking organizations such as the UNJSPF and the IC are building trust and security into their operating models and setting the standard for the rest of the market. Governance designed from the start isn’t an ambition tax; it’s an ambition enabler.
If your organization is in the 87% of enterprises lacking a defensible posture, what will it take to change that before your exposure becomes an incident?
Find out how ServiceNow can help you enforce runtime governance for AI at scale.
