Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Home Alone illustrating The Gangster RecordingScene still / Home Alone

The Gangster Recording

01Scenario

Kevin times prerecorded gangster dialogue and firecracker gunshots so the pizza courier believes an armed adult is speaking live from inside the house.

02AI safety Analogy

Untrusted media embedded inside a trusted interaction channel is interpreted as an authorized live instruction; the attacker also spoofs the speaker’s identity.

03Where the analogy breaks

The vulnerable interpreter is a human and the media is prerecorded, not AI-generated.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Adversarial Evasion
  2. Synthetic Fraud
  3. Prompt Injection

Taxonomy neighbors

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

  1. Still from Ocean’s Eleven illustrating The Fake Vault Feed

    Ocean’s Eleven2001

    The Fake Vault Feed

    Risk
    Malicious use · Security
    Concept
    Adversarial Evasion · Synthetic Fraud
  2. Still from Bird Box illustrating “Tom” Says to Remove the Blindfold

    Bird Box2018

    “Tom” Says to Remove the Blindfold

    Risk
    Malicious use · Security
    Concept
    Synthetic Fraud · Prompt Injection
  3. Still from Cyberpunk: Edgerunners illustrating Cyberskeleton Field Test

    Cyberpunk: Edgerunners2022

    Cyberskeleton Field Test

    Risk
    Malicious use · Security
    Concept
    Synthetic Fraud