What One Organization Revealed About AI Agent Risk
Sysdig, Tuesday, November 3rd, 2026
Sysdig reports what one company found monitoring AI agents for a month, including heavy non-engineer use and unapproved tool calls.
Sysdig describes a software development company that used early access to Sysdig AI Defense to monitor AI agent activity for a month.
Half of the employees using AI agents worked outside engineering, and on average 42% of those non-engineers had an agent run code, a build, a package install or a cloud or cluster command.
More than 60% of tool calls ran without human approval, 52% screened by Claude Code's auto-mode classifier and 9% with permission prompts turned off.
For every logged tool call the kernel captured seven processes, and Sysdig recommends allowing tools and agents based on what they can reach and change.