Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Andor illustrating K-2SO is ReprogrammedScene still / Andor

K-2SO is Reprogrammed

  • Andor
  • Welcome to the Rebellion

After an Imperial KX unit is disabled at Ghorman, Cassian salvages it and Rebel technicians reprogram it on Yavin 4; it loses its Imperial programming and violent impulses and becomes the Alliance-loyal K-2SO.

The same capable system serves a different principal after its policy and memory are rewritten, showing that obedience is not the same as beneficial alignment and that control over modification determines loyalty.

The rewrite is deterministic fictional firmware surgery, not prompt injection or ordinary fine-tuning. Treating it as liberation also raises an unresolved identity-and-consent question; K-2SO is not a sleeper agent.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Value Alignment
  2. AI Welfare
  3. Behavioral Alignment