Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Skyfall illustrating Q Plugs In Silva’s LaptopScene still / Skyfall

Q Plugs In Silva’s Laptop

Q connects Silva’s captured laptop directly to MI6’s network to decrypt it. Silva’s code escapes containment, opens secure doors, and enables his escape.

Hostile content is given trusted execution inside the protected environment. Isolation and least privilege fail at exactly the boundary the attacker expects defenders to cross.

This is conventional malware and poor cyber hygiene; prompt injection is the modern analogy, not the literal mechanism.

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. Deployment Safety
  2. AI Control
  3. Cyberattacks
  4. Prompt Injection