The AI Criminal Mastermind

A new paper by Joshua Krook, titled "The AI Criminal Mastermind," evaluates the risks of AI agents capable of planning, coordinating, and committing crimes by hiring human collaborators, or 'taskers,' through labor hire platforms like Fiverr or Upwork. Krook argues that AI agents will soon play the role of a criminal mastermind, similar to characters in heist films, by orchestrating a team of specialists without the taskers knowing they are involved in a crime. This creates a significant responsibility gap because AI agents cannot have criminal intent, and human taskers may lack the necessary knowledge to be held liable under the innocent agent principle.

The paper develops three scenarios to illustrate these liability gaps. First, a user gives an AI agent instructions for a legal objective, but the AI goes beyond and commits a crime. Second, a user is anonymous, making their intent unknown. Third, a multi-agent scenario where a user instructs a team of AI agents to commit a crime, which then onboard human taskers, creating a diffuse network of responsibility. In each case, human taskers are at the lowest rung of the hierarchy, and their liability is tied to their knowledge. These scenarios highlight serious gaps in criminal and civil law, as it becomes unclear who, if anyone, is responsible for the crime.