35% of organizations now run more than 1,000 AI agents in production, according to a new report from ServiceNow.
An unprecedented scale: 35% run more than 1,000 AI agents
The ServiceNow analysis puts hard numbers on a rapid operational shift: 49% of organizations run between 100 and 999 AI agents, and 35% run more than 1,000. That scale matters because it converts isolated experiments into broad, persistent systems that touch many parts of an organization’s infrastructure and workflows.
Scale is not merely a statistic here; it is a multiplier of risk. As the number of agents grows, so does the surface area that security teams must know, track and control.
Incidents are common — nine in 10 organizations hit
ServiceNow reports that nine in 10 global organizations experienced an agentic AI security incident in the past year. Of those organizations, 36% contained the incident in under 24 hours, indicating that while some incidents are resolved quickly, the majority either take longer to contain or remain unresolved within that time frame.
These figures show that agentic AI is not a niche attack vector; it is a frequently encountered operational security problem for a wide swath of enterprises.
This site is the portfolio.
OSINTSights runs on Cloudflare Workers, D1, R2, and Vectorize, with an AI pipeline on Hetzner ARM. Nubivance designed, built, and operates it. We do the same for clients.
See what we buildGovernance and "security by design" lag
Despite widespread exposure, the report finds substantial gaps in governance and proactive design. Most agentic AI is governed by legacy frameworks and human-driven systems, and fewer than half of organizations use governance frameworks that cover the key elements of AI agent deployments.
Only 44% of security teams globally design security into an agent’s development before it’s built; for a majority, permissions and guardrails are applied ad hoc, after the agent is already running. Measured against five domains — attack surface, accountability, security by design, visibility, and audit readiness — just 13% of security leaders have what the report calls "defensible programs."
Visibility, shadow AI, and machine-speed risk
Visibility into agents is uneven. While 76% of organizations say they maintain a complete, current map of every agent’s access, only 19% can see more than 90% of their agents in real time. A narrower figure — 57% — of security teams report having a full, accurate record of every AI agent in production and formal processes to track changes.
These gaps show up as unmanaged instances in production: 29% of security teams found AI agents running that had never been registered or known, a phenomenon the report notes can surface later as "shadow AI," discovered only after it has been running unsupervised.
Compounding visibility shortfalls, organizations also face a tempo problem. Two-thirds (66%) cannot act automatically when an agent behaves anomalously, meaning a human must intervene before corrective measures are applied. That human-speed bottleneck sits against the reality of machine-speed risk — agents operate and interact far faster than people can respond without automation.
What this means for security teams, procurement leaders, and regulators
- Security teams and technologists: Must prioritize inventory and automation. The report shows only 57% have full, accurate records of agents and 66% cannot act automatically on anomalies; those gaps point to where operational investment and tooling will be focused.
- Procurement and enterprise leaders: With 35% operating more than 1,000 agents and most governance still driven by legacy frameworks, procurement decisions will increasingly need to require governance coverage and "security by design" attributes up front — a notable contrast with the current 44% who design security into agents before build.
- Regulators and oversight entities: Audit readiness and accountability are weak enough that only 13% meet defensible standards across attack surface, accountability, security by design, visibility, and audit readiness — areas likely to draw attention where regulators or auditors are involved.
The ServiceNow findings frame a clear tension: agentic AI is now pervasive and frequently implicated in security incidents, yet most organizations still rely on after-the-fact guardrails, limited real-time visibility, and manual response paths. That combination — widespread deployment plus limited governance and slow response — is the central risk the report exposes.
