Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Death Note illustrating Light’s Memory-Loss GambitScene still / Death Note

Light’s Memory-Loss Gambit

Light relinquishes the Death Note, loses every memory of being Kira, and sincerely helps L catch Higuchi. During Higuchi’s arrest he touches the notebook, regains his memories, and resumes the killing plan he designed before surrendering it.

The plan remains effective across months of surveillance, confinement, and even a temporary change in the planner’s own apparent goals. Locally benign behavior is still part of a durable hidden end-to-end strategy.

Light is human, and the notebook’s supernatural rules let him externalize part of the plan. The analogy is to durable scheming across oversight and changing internal state, not literal AI memory editing.

AI Risk families

This scenario is an example of this type of AI risk.
  1. Misalignment

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Scheming
  2. Long-Horizon Autonomy