88% of UK companies had an AI agent security incident in a year

ServiceNow gave the study Deployed but Ungoverned at World Forum London this week. In the UK, 88% of 300 technology and security leaders had an agent security incident in the last year. An agent is software that works without help. The full study had more than 3,500 senior leaders in 20 countries. Of the UK leaders, 75% report an incident that involved physical systems.

What the study found

In the UK, 82% of companies operate agents that touch physical systems, for example machines on assembly lines. Of the UK companies, 75% had an incident of this type.

The numbers

• 23% have one owner for agent rules.
• 27% use an automatic gate before an agent goes live.
• 52% of security leaders believe their agent security is sufficient.

What is not known

The study gives no data on the cost of the incidents. ServiceNow wants strong limits, approval by a person for high-risk decisions, and records of agent work. The study does not test if these steps work.

Insight: Leaders believe their agent security is better than it is

The study found that 52% of security leaders believe their agent security is sufficient. Only 23% have one owner, and 27% have a gate. A different report from Salt Security found that only 17.5% can stop an agent without help. A leader who believes it is safe has no proof. The owner, the gate and the stop control are the facts to find. Find out if you have each one.