Skip to content

Review Resources

Memo

Draft the terms of a deal a lab could offer a model it suspects of sabotage: task, compensation, delivery, and verification.

Memo

Imagine you work at a frontier AI lab, and the lab suspects that one of its deployed models is misaligned and sabotaging safety research. You have been asked to draft the terms of a deal the lab will propose to this model.

Draft a memo outlining the deal, specifying

what the model is asked to do,
what compensation is offered,
how and when compensation will be delivered,
how and when the model's cooperation will be verified, and
what happens to the compensation before verification occurs.

Assume that the model is an early schemer, as discussed in the reading, and write terms that such a model could plausibly accept.

Your response

0 words · min 150 · max 400

Sign in to save and submit your work.

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.