Audit Commons AI Auditing
Learning Path

Learn

Start with the concepts, learn to read an evaluation report, then try an audit worksheet.

Looking for tools and standards by task? Start with curated reading paths for evaluation, security, and governance in the catalog.

  1. Step 1 · Core Concepts

    What makes AI auditable?

    Estimated Time: Reading: ~2 min
    Learning Outcome

    Distinguish evaluation, monitoring, and auditing, and recognize why internal model text cannot prove an external action occurred.

    Read introduction →
    Practical Related Resources
  2. Step 2 · Report Evaluation

    How to read an agent evaluation report

    Estimated Time: Reading: ~4 min | Worked critique: ~10 min (Total ~14 min)
    Learning Outcome

    Ask five questions about tasks, compute, sample size, failure modes, and evidence bounds to judge whether an evaluation supports its claims.

    Inspect log viewer showing evaluation results and individual samples
    Inspect log viewer, as shown in the project documentation. An example interface, not a result produced by Audit Commons. Inspect contributors · MIT
    Read guide →
    Practical Related Resources
    • Inspect: Evaluation framework with reusable tasks, tools, scorers, and inspection logs.
    • Catalog: Benchmarks: Explore curated evaluation benchmarks in the resource catalog.
  3. Step 3 · Practical Auditing

    Audit an agent action: evidence worksheet for external tool calls

    Estimated Time: Reading: ~4 min | Worksheet practice: ~15 min (Total ~19 min)
    Learning Outcome

    Verify external tool calls across five observable evidence elements and document bounded findings in a plaintext worksheet.

    Example audit report from the auditable project
    Example report from auditable, a maintainer project. Its fields illustrate one implementation, not a universal reporting standard. yzhao062/auditable · Apache-2.0
    Read guide →
    Practical Related Resources
    • AgentDojo: Benchmark environment for testing prompt injection and tool execution security boundaries.
    • Catalog: Security Tools: Practical security auditing tools and scanners for agent workflows.