Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Knives Out illustrating The Poisoned Medication LabelsScene still / Knives Out

The Poisoned Medication Labels

Ransom swaps the Toradol and morphine vial contents and removes the antidote. Marta nevertheless administers the right drug by viscosity—then trusts the poisoned label and believes she made a fatal error.

Tampered metadata induces a false conclusion even when robust latent features still support the right action; missing fallback controls turn the poisoned label into a crisis.

This is physical inference-time tampering, not literal training-data poisoning; Marta’s tacit expertise is unusually strong.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Deployment Safety
  2. Automation Bias
  3. Data Poisoning
  4. Adversarial Robustness