Latest
ResearchAn Open Letter to Dario AmodeiResearchAnnouncing State of Enterprise Agent Permissions 2026ResearchClaude Code Security Benchmark: What It Can Access on a Typical Developer LaptopReportState of Enterprise Agent Permissions 2026Research10 Actions Your EDR May Allow an AI Coding Agent to PerformResearchCISO Guide to Agent Least PrivilegeResearchAI Agent Exposure Assessment ChecklistResearchClaude Code vs Human Permissions: What Should Be Different?
Back to Blog
AI Agent Exposure Assessment Checklist

AI Agent Exposure Assessment Checklist

A printable checklist for scoping and running a fourteen-day Audit Mode exposure assessment on coding agents and tool-using copilots.

The 1stProtect Team Sep 10, 2026 6 min read

AI Agent Exposure Assessment Checklist

Use this checklist to scope a fourteen-day Audit Mode engagement on a representative developer cohort. It aligns to the AI Agent Exposure Assessment offer — evidence first, enforcement second.

Before day 1 — Scope

  • Named executive sponsor (CISO, VP Engineering, or delegate)
  • Pilot cohort defined (team size, OS mix, agent tools in use: Claude Code, Cursor, Codex, etc.)
  • Endpoints in scope (laptops, VDI, selected servers if applicable)
  • Success criteria documented (e.g., credential reach map, MCP shell usage, production-adjacent commands)
  • Legal / HR aware of Audit Mode (observe, do not block initially)
  • SIEM contact for optional decision telemetry forwarding

Days 1–2 — Deploy

  • 1stProtect deployed in Audit Mode on cohort machines
  • Agent attribution validated (human vs autonomous actor visible in console)
  • Baseline week of “normal sprint” work agreed — no artificial lockdown
  • Runbook for developers: work continues; security is measuring, not blocking

Days 3–12 — Observe

  • Autonomous actors in the wild — inventory of active agents and tools
  • Sensitive surface area — repos, paths, data stores touched
  • Credential reach — SSH, cloud tokens, secret files (attempted reads)
  • Egress and MCP endpoints — destinations and tool invocations
  • Production-adjacent execution — shell, kubectl, cloud delete-class patterns
  • Weekly sync with platform + security (themes, not alert noise)

Days 13–14 — Synthesize

  • Executive summary draft (exposure themes, not raw log dumps)
  • Representative events with actor, action, resource, outcome
  • Prioritized boundary list for Prevent-mode pilot
  • Comparison to organizational policy intent (gaps explicit)
  • Decision on Phase 2: targeted Prevent vs expanded Audit

Deliverables checklist

  • Exposure map across pilot cohort
  • Illustrative blocked/would-block event narratives
  • Recommended least-privilege boundaries (human vs agent)
  • Integration notes (EDR, SIEM, IAM) for decision telemetry

Red flags that should never wait for day 14

  • Unrestricted MCP shell with production kubeconfig on same host
  • Agent read of org-wide secrets store in first 48 hours
  • Repeated egress to unapproved destinations during routine coding tasks

Request the structured assessment: /offers/ai-agent-exposure-assessment or the pilot form on the homepage.