Topics

How advanced AI could let a small group gain or entrench outsized control, and how to prevent it.

• AI-enabled coups 1
• Economic concentration 1
• Surveillance and state power 2
• Checks and balances 1

Understanding what models compute internally.

• Mechanistic interpretability 2
• Probing and representations 1

Measuring dangerous capabilities and propensities.

• Dangerous capability evals 1
• Scheming and deception 2

Keeping untrusted models safe to use.

• AI control 1
• Scalable oversight 1

• Agent foundations 1
• Value learning 1

• Compute governance 2
• International coordination 1
• Lab policy and standards 1

Protecting model weights and infrastructure.