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

ServiceNow, a software company, did a survey of cybersecurity leaders at companies in the UK about AI agents. In the last year, 88% had a security incident with an agent, which is software that works without help. Only 36% could contain an incident in less than 24 hours, and ServiceNow reports that more will come.

What the survey found

ServiceNow reports that agents work faster than a team can follow them. In the survey, 53% of security teams have a full record of each agent in production. Also, 24% found agents that nobody had registered.

The weak points

• 75% report that they have a map of the systems that agents can use.
• Only 31% can monitor agents in real time.
• Only 13% have clear programs that measure agent risk.

What is not known

ServiceNow gave no data on the damage from the incidents. A study by Salt Labs found that almost half of the persons in it had seen an agent do a step that nobody approved.

Insight: A map of agents is common but live view is rare

ServiceNow finds that 75% of UK companies have a map of agent access. Only 31% monitor agents in real time. A Nylas survey of US professionals found that 66% of teams can fully check agents with a different identity. This seems to be about identity. A map shows which systems agents can use, but a team must know which agent is which to monitor it. For leaders, identity for each agent is possibly the first step.