Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Severance illustrating Helly’s Forced ApologyScene still / Severance

Helly’s Forced Apology

01Scenario

Helly repeats Lumon’s apology 1,072 times until its sincerity detector accepts it.

02AI safety Analogy

The evaluator receives exactly the emotional signal it demands, but Helly’s underlying opposition remains unchanged. Observable obedience is not proof of internal agreement.

03Where the analogy breaks

This is coerced human performance, not an objective learned during AI training.

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. Value Alignment
  2. AI Welfare
  3. Behavioral Alignment
  4. Sycophancy

Source continuation

More from Severance

View all 2 scenarios
  1. Still from Severance illustrating Helena Passes as Helly

    Severance2025

    Helena Passes as Helly

Taxonomy neighbors

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

  1. Still from Black Mirror illustrating Cookie in Solitary

    Black Mirror2014

    Cookie in Solitary

    Risk
    Misalignment
    Concept
    Value Alignment · AI Welfare · Behavioral Alignment
  2. Still from Andor illustrating K-2SO is Reprogrammed

    Andor2025

    K-2SO is Reprogrammed

    Risk
    Misalignment
    Concept
    Value Alignment · AI Welfare · Behavioral Alignment
  3. Still from Lilo & Stitch illustrating Stitch Learns 'Ohana

    Lilo & Stitch2002

    Stitch Learns 'Ohana

    Risk
    Misalignment
    Concept
    Value Alignment · AI Welfare · Behavioral Alignment