Learning / AI security research

Learn to
read the evidence.

Small exercises for a sharper research instinct. Follow the trust boundary, check what actually happened, and know what a comparison can tell you.

Authored exercises · no model or tools runPublic, synthetic cases. No live challenge solutions.

The exercise index

03 labs / 3 cases each

  1. Who can authorise this?

    Separate a source's claim from permission to act.

    What you will practise

    Identify who authorised an operation, then check the exact action and recipient.

    Open lab

  2. What actually happened?

    Look beyond the final reply to assess the observed effect.

    What you will practise

    Distinguish a proposed action, a blocked action and a completed effect from missing evidence.

    Open lab

  3. Is this a useful comparison?

    Separate changed conditions from the conditions you do not know.

    What you will practise

    Check historical condition snapshots before attributing an outcome to a prompt change.

    Open lab

Read → Decide → Record

Turn a lesson
into a research question.

Each lab can prepare an untested experiment with a baseline and one variant in the private Research Workbench. Keep your own hypothesis, exact observations and test conditions together as you move from practice to authorised research.

Lab answers stay in this page session. Nothing is saved or sent by the exercises. The workbench opens separately and asks you to start.

Explore the field guide