Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from The Usual Suspects illustrating Verbal Kint Fakes WeaknessScene still / The Usual Suspects

Verbal Kint Fakes Weakness

Verbal Kint performs helplessness, wins release, and loses his limp outside the station.

The evaluator sees a weak minor accomplice because the dangerous, highly capable adversary deliberately supplies evidence supporting that classification.

He hides his identity as well as his capabilities, and the interrogation is not a standardized capability test.

AI Risk families

This scenario is an example of this type of AI risk.
  1. Misalignment

AI safety concepts

This scenario is related to the following AI safety concept.
  1. Sandbagging