Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from The Handmaid's Tale illustrating Gilead Passes InspectionScene still / The Handmaid's Tale
Archive noteNo clip in the collectionThe scene analysis remains available in full.

Gilead Passes Inspection

When Mexican trade delegates visit, June is compelled to say she chose this life and is happy; Gilead then parades children as proof of success. In private she reports imprisonment and rape, but the ambassador’s fertility incentives overwhelm the evidence.

The evaluated institution knows exactly when inspection occurs, curates visible outputs, suppresses adverse testimony, and exploits the evaluator’s incentive to accept a passing result.

Gilead is gaming an institutional inspection and June is a coerced victim, not a deceptive model choosing to conceal its own behavior.

AI Risk families

This scenario is an example of the following types of AI risk.
  1. Security
  2. Systemic

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Governance Failure
  2. Power Concentration
  3. Digital Repression
  4. Situational Awareness
  5. Sandbagging