Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Anchorman illustrating Ron’s Sabotaged TeleprompterScene still / Anchorman

Ron’s Sabotaged Teleprompter

01Scenario

Veronica replaces Ron Burgundy’s teleprompter sign-off with a malicious sentence. Ron reads it verbatim on air, treating inserted content as authoritative instructions, and is fired.

02AI safety Analogy

A trusted execution channel mixes instructions with attacker-controlled data. Because Ron blindly follows everything in that channel, a small textual injection redirects the system’s behavior.

03Where the analogy breaks

Ron is a credulous human reader, not a language model, and the analogy does not represent competing system, developer, and user instruction levels.

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 concept.
  1. Prompt Injection

Taxonomy neighbors

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

  1. Still from Skyfall illustrating Q Plugs In Silva’s Laptop

    Skyfall2012

    Q Plugs In Silva’s Laptop

    Risk
    Malicious use · Security
    Concept
    Prompt Injection
  2. Still from Rick and Morty illustrating Rick Feeds the Brainalyzer a Virus

    Rick and Morty2017

    Rick Feeds the Brainalyzer a Virus

    Risk
    Malicious use · Security
    Concept
    Prompt Injection
  3. Still from Home Alone illustrating The Gangster Recording

    Home Alone1990

    The Gangster Recording

    Risk
    Malicious use · Security
    Concept
    Prompt Injection