Introduction, overview, and threat modeling

Before the session

• Skim all six module readings yourself
• If you run a classroom, open its dashboard and check each student's progress column for this module
• Check the submissions column for completed written responses; pick one or two anonymized excerpts you can read out as discussion starters (ask permission).
• Have the Control scenarios exercise open on a screen you can share.

Agenda

Times are a running clock for a 60-minute session

Warm-up: what is control? (0:00–0:10)

Have two or three participants state the core bet of the case for controlling powerful AIs in one sentence each, without looking. Write the phrasings on the whiteboard and let the group edit them into one agreed sentence; leave it up all session.

Prompt:

• "What differentiates alignment and control"

The control game, together (0:10–0:30)

Walk the setup of the AI Control paper: who the red team is, who the blue team is, what winning means. Then run Control scenarios on the shared screen — per scenario: read aloud, hand-count vote, one majority voice and one minority voice defend, then reveal.

Prompt:

• "Which assumption of the game felt least realistic to you, and what would change if you dropped it?"
• If a vote is unanimous: before revealing, make the room construct the strongest case for the answer nobody picked.

Catching and consequences (0:30–0:40)

In pairs, answer: "According to Catching AIs red-handed, what changes after the first catch?" Then compare answers across pairs; if the group wants a structured recap, run why catching counts on the shared screen.

Prompts:

• "Who needs to be convinced by a catch, and what evidence would actually convince them?"
• "What does a false alarm do?"

Threat prioritization debate (0:40–0:55)

Draw the concentrated-versus-diffuse columns from Prioritizing threats for AI control on the whiteboard. The group places example failures from the readings into a column; disputed placements get argued, one minute per side.

Prompts:

• "Which column does this failure go in?"
• "Are there failures that are ambiguous, what makes it this way?"

Wrap-up (0:55–1:00)

Each participant names one question they still have, in one sentence, no discussion. Collect these; they are your agenda seeds for the next session and good prompts for the module's remaining exercises.

Prompt: "What is one question this module raised that the next module had better answer?"

After the session

• On the classroom dashboard, check whether participants who attended go on to complete the module's remaining readings during the week.
• Read new submissions for the module's written exercises; recurring confusions belong on next session's warm-up.
• If several participants asked the same wrap-up question, post it (and where the readings address it) in your group's channel rather than waiting a week.