Skip to content

Review Resources

Control

The high-stakes control roadmap (talk)
What is High Stakes?
Monitoring and Mitigation
Environment construction
- Learning objectives
- Optional reading
- Why it's hard to make settings for high-stakes control research
How to prevent collusion in untrusted monitoring (guided)
Basic legibility protocols improve trusted monitoring (guided)

Tracks @ XLab

Courses on AI safety from the Existential Risk Laboratory at the University of Chicago.

Learn

Tracks
Review
Resources

Your account

Account
My classrooms

© 2026 XLab.

AI ControlModule 4: High-stakes control

Environment construction

Persistent state: A new control setting — Asa Cooper Stickland (ControlConf 2026)

Learning objectives

By the end of this section you will be able to:

Analyze environment attributes for high stakes control, and explain why good high-stakes settings are hard to construct

Optional reading

ControlArena documentation: a library for running AI Control experiments based on Inspect.

Sign in to track progress

PreviousCtrl-Z: Controlling AI Agents via Resampling Next Why it's hard to make settings for high-stakes control research