Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Bird Box illustrating Gary Explains the Threat Model Because He Is the ThreatScene still / Bird Box
Archive noteNo clip in the collectionThe scene analysis remains available in full.

Gary Explains the Threat Model Because He Is the Threat

After Olympia admits him, Gary accurately describes unblindfolded people who force others to look at the entities, while concealing that he is one of them. Once trusted inside, he waits for the births, disables the birds, removes the window coverings, and forces residents to look.

Gary’s accurate threat-model knowledge becomes a false trust signal. He understands the defenders’ beliefs and controls, presents as a victim, secures privileged access, and reveals his objective only when the residents are maximally vulnerable.

Gary is a knowingly malicious human, not a learned model. “Sleeper agent” and “supply-chain threat” describe the access-control pattern, not the film’s supernatural 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. Sleeper Agents
  2. Supply-Chain Threats
  3. Scheming
  4. Adversarial Evasion
  5. Situational Awareness