Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Iron Man 2 illustrating Tony Hijacks His Own Oversight HearingScene still / Iron Man 2

Tony Hijacks His Own Oversight Hearing

Called before the Senate to surrender the Iron Man armor, Tony takes over the hearing-room displays from his phone, plays footage of failed foreign and Hammer prototypes, and declares that he has “privatized world peace.”

The regulated actor controls the regulator’s information channel, curates the evidence, and turns a technical capability gap into political leverage. Oversight is not independent when its subject can compromise its tools or monopolize the expertise needed to interpret the evidence.

Tony’s evidence is substantially truthful, and he alters neither a learned reward nor a formal evaluation score. This is evaluation-channel security and procedural capture—not wireheading, prompt injection, or AI misalignment.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Cyberattacks
  2. Governance Failure
  3. Power Concentration