Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Memento illustrating Poisoning Future MemoryScene still / Memento
Archive noteNo clip in the collectionThe scene analysis remains available in full.

Poisoning Future Memory

01Scenario

Leonard deliberately records Teddy's license plate as a new tattoo fact and marks Teddy as untrustworthy, knowing his future self will treat those artifacts as objective evidence and kill him.

02AI safety Analogy

An agent writes false information into its own persistent context so that a later version of itself will pursue a goal the current version selected but cannot retain.

03Where the analogy breaks

Teddy's full account remains uncertain, but Leonard's deliberate falsification of evidence for his future self is explicit.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Reward Hacking
  2. Goal Misgeneralization
  3. Data Poisoning
  4. Long-Horizon Autonomy

Taxonomy neighbors

Selected from other sources by shared risk families and AI safety concepts.

  1. Still from Attack on Titan illustrating Eren’s Memory Poisoning

    Attack on Titan2022

    Eren’s Memory Poisoning

    Risk
    Misalignment · Malicious use · Security
    Concept
    Data Poisoning · Long-Horizon Autonomy
  2. Still from True Detective illustrating Poisoning the Audit Trail

    True Detective2014

    Poisoning the Audit Trail

    Risk
    Malicious use · Security
    Concept
    Reward Hacking · Data Poisoning
  3. Still from Inception illustrating The Idea Won’t Wake Up

    Inception2010

    The Idea Won’t Wake Up

    Risk
    Misalignment
    Concept
    Goal Misgeneralization · Data Poisoning