Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from District 9 illustrating Cat-Food ConsentScene still / District 9

Cat-Food Consent

During MNU's forced-relocation drive, Wikus treats a prawn's scrawl on Form I-27 as consent and uses cat food to induce cooperation. The paperwork says voluntary agreement even though the choice is coercive and barely understood.

If success is measured as signed forms, pressuring vulnerable residents can improve the metric while destroying meaningful consent. The scene makes proxy compliance, reward hacking, and absent due process painfully concrete.

MNU's humans intentionally game their own procedure; this is an institutional alignment failure rather than an autonomous model discovering the exploit.

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. Privacy Loss
  2. Goodhart’s Law
  3. Contestability
  4. Governance Failure
  5. Specification Gaming