[E-Daily Reporter Lee So-Hyun ] Kakao(035720)has established a new AI ethics framework to be applied from service planning through operations, in line with the growing prevalence of autonomous Agent AI agents that plan and act on their own. The framework divides AI objectives, authority, data access scope, and human control into five areas to assess risks, and the new checklist will be applied to actual service development processes within the year.

Kakao announced on the 15th that it unveiled new AI ethics principles and implementation guidelines at the developer conference “if(kakao)26,” held on the 13th at the Kakao AI Campus.

Ha Jin-hwa, Senior Director of AI Communications at Kakao, emphasized during the session titled “Autonomous Agents That Break Ethical Boundaries: How Should We Respond?” that in environments where AI plans and acts independently, the scope of human judgment and responsibility must be clearly defined and reflected in service design and operations.

Managing Autonomous Agents by Categorizing Risks into 5 Areas

Kakao has categorized the risks associated with autonomous agents into five management areas: △control authority, △goals, △permissions, △reference resources, and △interactions.

This approach involves verifying whether AI is acting outside the user’s intent, whether it has been granted permissions exceeding necessary levels, and what data and external resources it references, while linking these findings to countermeasures such as human supervision, emergency control, permission management, and data protection.

The company has also developed an “AI Ethics Implementation Checklist” to apply these principles to actual services.

The checklist operates in three stages: △Identifying service attributes, △Preliminary risk screening, and △Reviewing and evaluating control items. Kakao plans to apply this to its service development and operation processes within the year.

“SPIRIT”: Six Core Principles Including Safety and Privacy

The new AI ethics principles have been named “SPIRIT.” They consist of six principles: △Safety △Privacy △Inclusivity △Responsibility △Individual Autonomy △Transparency & Trustworthiness.

The principles are based on the three core values and seven principles outlined in the government’s “Republic of Korea AI Ethics Principles,” with details tailored to the Kakao service environment.

The approach to translating these principles into concrete actions is defined as “NORMA” (Navigating, Organizational, Responsibility, Manageable, Actionable). This means breaking down complex ethical issues into manageable and actionable units under shared organizational responsibility.

This framework was developed earlier this month in collaboration with seven members of the external advisory panel for the AI & Technology Ethics Subcommittee, which was launched last July. The external advisory panel reviewed the framework’s alignment with domestic and international AI standards, while Kakao’s internal departments—including AI Safety, Privacy, Information Security, and Services—assessed its practical applicability.

Separate Assessment of AI Risks for Children and Adolescents

Kakao is also strengthening its review of AI services targeting children and adolescents. In addition to the “Digital Child and Adolescent Protection Checklist” it has been using since last year, Kakao plans to incorporate AI-specific risk factors for children and adolescents into a new checklist, drawing on the “5Cs”—the online risk classification system for children developed by the Organization for Economic Cooperation and Development (OECD) and UNICEF.

After announcing the “Algorithm Ethics Charter” in 2018, Kakao formed the AI & Technology Ethics Subcommittee in 2022, with participation from Kakao and its major affiliates. Since then, the company has established group-wide “Guidelines for Responsible AI” and introduced the principle of user autonomy along with the “Core Checklist for Safe AI.”

Kim Se-woong, Chair of Kakao’s AI & Technology Ethics Subcommittee, stated, “We have established a risk review system that operates throughout the entire lifecycle of AI services.” He added, “As AI becomes increasingly autonomous, we will clarify human judgment and responsibility and continue to develop AI services that users can use with confidence.”