Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Independence Day illustrating Independence Day VirusScene still / Independence Day

Independence Day Virus

Levinson uploads a virus to the alien mothership. Because its commands propagate to every destroyer, the virus disables the entire fleet's shields at once and opens a brief window for a global counterattack.

A homogeneous fleet inherits one shared vulnerability, so one upstream compromise becomes a system-wide correlated failure. It is a famous illustration of the danger in deploying the same model, control plane, or safety layer everywhere.

Human malware running on an alien operating system is famously implausible—the comic part of an otherwise useful architecture lesson.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Deployment Safety
  2. Infrastructure Dependence
  3. Algorithmic Monoculture
  4. Cyberattacks
  5. Adversarial Robustness