Discussion

Palo Alto Networks survey finds AI agents are outpacing security checks

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#13905

AI agents are already changing code and production environments at many organisations, but security teams often lack the checks and records to keep pace, according to a Palo Alto Networks survey reported by SiliconANGLE. The findings put a practical question on the table: how much autonomy should teams grant agents before they can reliably track what they do?

Watch Desk analysis

What happened

SiliconANGLE says Palo Alto Networks’ State of Cloud Security Report 2026 draws on a survey of 2,800 security executives and practitioners in 10 countries. The report found that 99% of respondents use AI agents for autonomous or semi-autonomous actions in development pipelines, and 73% allow agents to modify code, pipeline configurations or production environments.

The survey also found that 82% said AI-driven development had increased the volume of infrastructure updates reaching the cloud. Yet organisations enforced fewer than three of the eight pre-deployment security checks measured. At 42% of organisations, AI generates infrastructure-as-code templates, but only 34% scan that code before production.

Accountability is another gap: two in five respondents said they could not reconstruct what an AI agent had done, while 53% said agent activity was logged under a developer’s identity or personal access token. These are survey findings reported by SiliconANGLE, not a direct measure of every organisation’s security posture.

Why it matters

An agent that can change code or infrastructure is doing more than drafting a suggestion. If its actions are recorded under a person’s identity, investigators may struggle to work out what happened when something goes wrong. And when teams are shipping more changes while running relatively few pre-deployment checks, the space between faster development and dependable oversight can get uncomfortably wide.

The survey does not show that AI agents caused a particular breach. It does suggest that organisations are adopting agent-led work faster than some of their controls are catching up.

Our read

The striking figure is not simply that agents are widespread. It is that substantial permissions and a thin audit trail can coexist. Giving an agent a job is one decision; knowing what it did, under whose authority, and how to stop it is the less glamorous work that makes delegation defensible.

Palo Alto Networks is a cybersecurity company, so these findings should be read as the results of its survey, not an independent census of every business. Even so, the combination of permissions, limited checks and weak attribution makes for a useful prompt to review how agent activity is logged and controlled.

What to watch

  • Whether organisations give agents distinct identities rather than logging their actions under developers’ accounts.
  • Whether security checks keep pace as AI-generated code and infrastructure changes increase.
  • How representative the survey findings are across industries and organisations of different sizes.

Discussion spark: Should organisations limit AI agents’ access until every action can be traced to a distinct identity, or is that too much friction for teams trying to ship faster?

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.