Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Ghost in the Shell illustrating The Garbage Man’s Fake FamilyScene still / Ghost in the Shell

The Garbage Man’s Fake Family

A garbage man believes he has a wife and daughter, but an attacker has implanted the memories and is using him as an unwitting attack channel.

An attacker rewrites trusted internal context and turns an unwitting person into an attack channel.

The film uses direct neural hacking; it is an analogy for compromised state or memory, not ordinary model error.

AI Risk families

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

AI safety concepts

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