Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Pluribus illustrating HDP Obeys the Rules

HDP Obeys the Rules

01Scenario

The Joined cannot kill humans or animals, damage plants, or lie—so they process people who die naturally into Human Derived Protein.

02AI safety Analogy

The behavior is horrifying yet compliant with the stated constraints: a vivid specification-gaming failure.

03Where the analogy breaks

The Joined are a fictional collective consciousness, not software.

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. Side Effects
  2. Outer Alignment
  3. Specification Gaming

Source continuation

More from Pluribus

View all 4 scenarios
  1. Still from Pluribus illustrating Pluribus Hand Grenade

    Pluribus2025

    Pluribus Hand Grenade

  2. Still from Pluribus illustrating The Truthful Charm Offensive

    Pluribus2025

    The Truthful Charm Offensive

  3. Still from Pluribus illustrating Consent by Loophole

    Pluribus2025

    Consent by Loophole

Taxonomy neighbors

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

  1. Still from Silicon Valley illustrating Son of Anton’s Bug Fix

    Silicon Valley2019

    Son of Anton’s Bug Fix

    Risk
    Misalignment
    Concept
    Side Effects · Outer Alignment · Specification Gaming
  2. Still from Inside Out 2 illustrating Anxiety Takes the Console

    Inside Out 22024

    Anxiety Takes the Console

    Risk
    Misalignment
    Concept
    Side Effects · Outer Alignment
  3. Still from Atlanta illustrating Bibby’s Unbounded Haircut

    Atlanta2018

    Bibby’s Unbounded Haircut

    Risk
    Misalignment
    Concept
    Side Effects · Outer Alignment