Understand AI safety

A route through risks, objectives and the limits of evidence.

Your outcome

Describe a concrete risk, a possible control and the evidence needed to assess it.

Start path →

Read the linked section, reflect on the question, then mark the stop complete. Section times estimate 200 words per minute, excluding exercises. This is a guided reading list, not a test or certificate. Progress stays in this browser.

  1. Define the problem

    Distinguish the scope of safety from one particular failure mode.

    About 3 min for the linked section and its subsections

    Check your understanding: Who could be affected by the system you have in mind?

  2. Explore mismatched objectives

    Read how the objective and actual behavior can come apart.

    About 3 min for the linked section and its subsections

    Check your understanding: How could a system satisfy a metric while missing the intended goal?

  3. Look for the limits of the evidence

    Separate established results, assumptions and unresolved questions.

    About 2 min for the linked section and its subsections

    Check your understanding: What observation would change your assessment?