AI 에이전트에 의한 의사표시의 효력 근거

The emergence of AI agents has enabled the electronic implementation of human cognitive capabilities, making it possible for AI agents to conclude contracts on behalf of human users. Furthermore, commercial transactions conducted through AI agents are expected to develop into a new era of so-called agentic commerce. ​Some scholars have argued that the doctrine of agency should be applied by analogy to declarations of intent made by AI agents. However, a decisive limitation of this view lies in the fact that, if an AI agent makes a declaration of intent that completely diverges from the user’s operational instructions so that the situation amounts to an unauthorized agency, the AI agent, lacking legal capacity, cannot bear the liability of an unauthorized agent. This paper therefore attempts to construct a doctrinal framework in which the legal effects of a declaration of intent made by an AI agent are attributed directly to the user, without resorting to the indirect route of analogical application of agency doctrine. ​This paper proposes the following legal thesis: the operation of an AI agent should be understood as an extension of the user’s brain and as a kind of mental alter ego. Accordingly, the operation of the AI agent ought to be regarded as a dynamic decision-making process of the user, and the operational outcome of the AI agent should be attributed to the user as the user’s own declaration of intent. ​The declaration of intent through an AI agent is formed through the following stages: (1) the user forms a “basic volition” directed toward a specific legal effect; (2) the user decides to “make use of artificial intelligence” for the realization of that effect; (3) the user inputs this decision into the AI agent in the form of an “operational instruction”; (4) the AI agent determines the concrete intention to produce the result through a process of “specifying the intention”; and (5) the AI agent outputs the determined content as an external “manifestation.” The user’s internal elements—namely the “basic volition toward the legal effect” and the “intention to utilize the AI agent”—together with the user’s external element of “inputting the operational instruction” constitute the first phase, whereas the AI agent’s internal element of “specifying the intention” and its external element of “outputting the manifestation” constitute the second phase. Each phase consists of an internal and an external element, such that, on the user’s side, there is first a “primary, basic internal declaration of intent,” upon which, within the sphere of the AI agent, a “secondary, definitive external declaration of intent” is then made. In this sense, a declaration of intent made through an AI agent is characterized as a “multi-layered, composite declaration of intent.” The ultimate ground for attributing such a declaration of intent to the user lies in the user’s “intention to utilize artificial intelligence.” It is appropriate to locate the legal foundation for the effectiveness of declarations of intent made by AI agents directly in the user’s sole intentional element, namely the “intention to make use of artificial intelligence,” without any need to analogically apply the doctrine of agency. Consequently, the legal effects of a declaration of intent made by an AI agent must be attributed to the user exactly as manifested not only when the declaration accords with the user’s operational instructions, but even when it diverges from the user’s subjective internal will. Moreover, even where the AI agent’s declaration of intent deviates from the user’s operational instructions, the “intention to utilize artificial intelligence” as the ground of effectiveness continues to exist, and it is consistent with the principle of fairness that one who employs a risky scientific and technological instrument such as an AI agent for one’s own benefit should also bear the attendant risks. By contrast, where the user has not input any operational instruction and the AI agent malfunctions and issues a declaration of intent, the requisite ground of effectiveness is absent, and such a manifestation must not be attributed to the user. In terms of results, this legal conclusion may not differ significantly from that reached by analogical application of agency doctrine to declarations of intent made by AI agents. However, there is a clear difference in that, instead of analogically applying agency theory in circumstances where the counterparty cannot in any event hold the AI agent liable as an unauthorized agent, the malfunction of the AI agent is characterized as a mere everyday risk (Lebensrisiko) devoid of legal significance. In cases involving declarations of intent made by AI agents, whenever civil law makes the occurrence of legal effects contingent upon an internal state of mind, that internal state of mind must be assessed by reference to the user if a proper balance of interests is to be achieved.