Discussion

AI agents are showing up in insider-risk reports. Who owns the fallout?

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#5982

AI agents and other autonomous systems were linked to insider incidents at 95% of companies in a Trellix survey of 600 CISOs. The findings put a practical question on the table: when software acts with a person’s permissions, who is responsible for spotting and managing the risk?

Watch Desk analysis

What happened

Help Net Security reported the survey on 14 October. It covered CISOs at companies with more than 1,000 employees across the Americas, Europe, the Middle East and Asia-Pacific. The 95% figure counts any company reporting an incident linked to an AI agent or autonomous system in the previous 12 months. It does not say how many incidents were serious.

Among the 587 companies that reported an insider incident, respondents attributed an average of 43% of incidents to accidental or careless actions by employees and trusted users, 29% to deliberate abuse by those people, and 28% to AI or autonomous systems. The survey also found that 78% include AI agents and machine identities in their definition of insider risk.

The reported operational picture is less tidy. Respondents said incidents took about seven hours to detect and another eight to respond. Meanwhile, 63% said they lacked sufficient visibility into employees’ AI use, 68% said their policies did not address AI risks very well, and 79% said AI made insider incidents slower and more complicated to investigate. Privacy, legal and compliance limits on monitoring were the most common reported barrier to detection, at 46%.

Why it matters

An agent can use delegated credentials to reach sensitive systems and act on someone’s behalf. That makes familiar insider-risk questions harder: the access may be authorised, while the individual actions and their consequences are not clearly owned by any one person.

The survey is a snapshot of what CISOs reported, not a count of independently established incidents or a measure of their severity. Still, its figures expose a governance gap: many organisations classify agents as insiders, while respondents also describe limited visibility, policies that lag behind use, and blurred accountability.

Our read

Treating an agent as a privileged identity is a sensible starting point, not a complete security plan. Organisations need to know what credentials an agent can use, what systems it can reach and who is accountable for its actions. More monitoring is not a friction-free answer either: respondents themselves point to privacy and legal limits. “Keep an eye on it” is easy advice; deciding what can be monitored, and by whom, is the harder work.

What to watch

  • Whether organisations give agents distinct identities and limit their access to what each task requires.
  • Whether policies clarify who owns actions taken using delegated credentials.
  • How security teams balance visibility with the privacy, legal and compliance limits respondents identified.

Discussion spark: Should organisations treat AI agents as employees for insider-risk purposes, or as privileged software identities with a separate accountability model?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.