Blog / Governance

95% of organisations have already had an AI agent incident, says Gartner

The survey Gartner released on 16 November puts a number on what many teams see without saying: agents already cause incidents, while most leaders believe they are under control.

By Vincent Dorange, , 2 min read

95%

of organisations surveyed faced at least one agent-related incident in 12 months

On 16 November 2026, as its London conference on infrastructure and operations opened, Gartner published the results of a survey of 298 infrastructure and operations leaders in the US and the UK, conducted in June and July 2026 1. The firm draws a projection from it: without strong governance, 40% of organisations scaling AI will suffer a business-critical service disruption by 2028.

Incidents already happen, confidence lags behind

The most striking figure looks backwards: 95% of the organisations surveyed faced at least one agent-related incident in the past 12 months. At the same time, 70% of leaders believe their agents currently operate within safe boundaries. Gartner calls this a massive operational readiness gap, and the survey spells it out. 57% of respondents say their current governance is weak, 55% find their IT risk tools completely outdated, and only 15% have automatic circuit breakers to stop runaway agents or infinite loops.

Seven pillars and an immediate instruction

For Nathan Hill, Distinguished VP Analyst at Gartner, leaders must immediately inventory active agents and register them as nonhuman identities in their identity platforms. The firm recommends a governance framework built on runtime controls and organised around 7 pillars 1:

  • identity: every agent is treated as a machine identity and tracked in a central registry so nothing runs in the shadows;
  • guardrails built into runtime, able to slow down a misbehaving agent before it disrupts operations;
  • observability of the reasoning chain, to understand why the agent did what it did;
  • cost control, with routing between models and session-level spending caps;
  • escalation based on risk level, with high-impact actions still requiring human sign-off;
  • containment, with identity kill switches to isolate runaway agents before failures cascade;
  • data perimeters against context poisoning and leaks.

Gartner adds a point worth more than a glance: the more an organisation depends on its agents, the less freedom it has to switch them off. A governance framework must therefore identify which agents have to keep running even when they are expensive or hard to maintain.

What the list takes for granted

These 7 pillars describe the technical control of an agent well. The escalation pillar, however, assumes a prior decision the survey does not address: who decided what counts as a high-impact action, and on what basis? The Agentic Commerce Framework® starts from that question. It asks for a written statement, decision by decision, of what the agent may do alone and what requires a human, and for a named person who answers for it. Runtime controls then enforce rules that have an author.

Gartner's last remark also calls for a simple precaution. A kill switch that has never been used proves nothing; stopping an agent that has become critical has to be prepared, with a fallback, well before the day it must be cut off.

Sources

Links to the official texts and original publications.

  1. 1Gartner Says Weak AI Governance Will Lead to Business-Critical Service Disruptions for 40% of I&O Organizations by 2028, Gartner, 16 November 2026.

Glossary terms used in this article

AI agent kill switch

Vincent Dorange, creator of the Agentic Commerce Framework®

Also worth reading

25

organisations interviewed; none lets its agents act with unrestricted autonomy

Governance

AI agents at work: what the OECD learned from 25 organisations

4 levers

transaction caps, budgets, approved counterparties and escalation rules, per Circle

Agentic economy

Agentic economy: agents that spend within delegated permissions

5 degrees

of autonomy in the CIANum scale the CNIL refers to

Personal data

Agentic AI: what France's data protection authority says about agents and personal data

All articles