WHO SHOULD CONTROL AI?
Governments?
Tech companies?
Or no one?
This week we look at how AI is actually governed today.
Not in theory, in reality.
Tree forces are shaping the rules right now:
- Governments trying to regulate AI
- Tech companies writing their own safety frameworks
- Incidents showing what goes wrong in practice
The unconformable truth:
AI governance is still being invented.
An nobody agrees on the rules.
SESSION OUTLINE
During this session we will:
1️⃣ Discuss real AI incidents
2️⃣ Examine frontier AI risks
3️⃣ Analyse current regulation (EU AI Act, SB-53)
4️⃣ Simulate an AI governance decision
The goal is simple:
Understand why governing AI is extremely difficult.
HOMEWORK
Please bring:
1 AI incident
1 plausible risk
1 concern about AI governance
1 opinion about current regulation
We will use these as the basis for discussion.
READING MATERIAL
📚 Pre-reading package — max 60-90 min total
1. Real AI Incidents (15 min)
OECD AI Incidents Monitor (watch + read 2-3 cases) Real incidents, no opinions. Source: OECD AI Incidents Monitor (database)
Assignment 1: Choose one incident and answer:
What went wrong?
Was it misuse, accident, or misalignment?
2. Frontier risks & governance lens (15-20 min)
International AI Safety Report (Extended Summary for Policymakers) This is literally designed to brief policymakers. Not academic material.
Source: International AI Safety Report (Extended Summary for Policymakers)
Assignment 2: Highlight 3 risks that you consider "already plausible."
3. Governance reality: company frameworks (20 min)
Scan one of these frameworks:
DeepMind Frontier Safety Framework (more accessible version)
Anthropic Responsible Scaling Policy (more accessible version)
Assignment 3: What capability threshold should stop deployment?
4. Regulation in practice (20 min)
EU AI Act overview — short explainer, not the entire law.
EU GPAI Code of Practice — a template for how companies can comply with the AI Act
California’s SB-53 (more accessible version) — most important piece of US legislation on general-purpose AI at the moment
Assignment 4: Write:
1 thing you like about current AI regulation
1 concern about its effectiveness