AISOF

Business risk, 2026 surveys

One in eight breaches now starts with an agent

Autonomous agents account for more than one in eight reported AI breaches, and among enterprises that deployed agents, the overwhelming majority report at least one incident.

AISOF Briefings, Web Dance Pty. Ltd2 sources
1 in 8: AI breaches now start with an agent

The 2026 survey picture is consistent across sources. Autonomous agents account for more than one in eight reported AI breaches. Among enterprises that have actually deployed agents, 88 percent report at least one security incident tied to them. Of organisations reporting agent driven incidents, 61 percent involved data exposure. IBM's Cost of a Data Breach research found one in four malicious breaches were AI enabled, at an average cost around 6 million US dollars.

Survey figures deserve the usual caution. Definitions of incident vary, respondents self select, and vendors commissioning research have an interest in the answer. Treat the precise percentages as indicative and the direction as reliable.

The direction is that agent incidents have stopped being anecdotes. When a majority of deploying organisations report one, the base rate has changed, and a board asking whether this could happen here is asking the wrong question. It is happening at peers, and the useful question is whether it would be detected.

That is answerable, and it is answerable now. Detection coverage over agent actions, retention long enough to reconstruct a session, and a named owner for every deployed agent are the difference between an incident found in two hours and one found in a customer's complaint.

Sources

Briefings summarise public reporting for educational purposes and are not legal advice. Editorial policy

Assessment packs

Request packs for the AI frameworks that apply to your organisation or product.