AI를 활용한 범죄컨설팅의 위험성 및 입법 필요성 연구

Professor Moriarty, who appears in the Sherlock Holmes novels, is Holmes’s eternal nemesis. Rather than committing crimes himself, Moriarty typically plays the role of assisting others who intend to commit crimes. Readers are drawn to Moriarty’s brilliance, as he offers strategic advice on criminal activities—from overarching plans down to intricate details—based on his genius-level intellect. Individuals like Moriarty are generally referred to as “crime coordinators,” and such actions are known as “crime consulting.” Typically, crime consulting involves one individual providing concrete advice to another on how to commit a specific crime. As such, crime coordinators are often well aware of the principal perpetrator’s intent and the acts to be carried out. Even if they do not fully grasp every specific action the principal plans to take, they generally foresee the legal interests that are likely to be infringed. Because crime coordinators recognize the crime being committed and actively assist in it, they are usually held liable as accessories to the principal offense. However, if the principal does not proceed to the execution phase, the crime coordinator can only be punished under legal provisions for preparation or conspiracy—provided such provisions exist. Otherwise, no punishment is possible. In other words, in order to hold a crime coordinator criminally liable, the principal must begin executing the crime, which reveals a structural limitation in relying on the principal’s actions. When it comes to conventional crime consulting—where one individual advises another—this legal interpretation is not particularly problematic, considering the risk and level of involvement. But applying this interpretation unchanged to crime coordinators who use AI as a tool may lead to a significant imbalance in punishment compared to the actual societal harm caused. This is because it is highly dangerous for an AI developer or manager to create a system capable of providing up-to-date and concrete methods for committing crimes. AI can provide crime consulting to a large number of potential offenders in a short period of time, potentially resulting in massive harm. This risk increases significantly if the AI is trained on the crime methods detailed in recent court rulings. This study conducted an experiment to determine whether an AI trained on specific fraud techniques found in recent judgments could provide malicious responses. As a result, it was confirmed that despite the presence of an ethical filter within the AI, it was still capable of generating such responses. This outcome demonstrates that it is technologically feasible to develop a malicious AI and use it for crime consulting purposes. Given the potential for AI-based crime consulting to cause widespread harm in a short period, it is imperative that the National Assembly promptly enact relevant criminal law provisions to protect the public from such threats.