Kakao Unveils 'SPIRIT' AI Ethics Framework and 'NORMA' Implementation System for Autonomous Agents
At if(kakao)26, Kakao announced 'SPIRIT', a set of six ethical principles tailored for autonomous AI agents, alongside 'NORMA', an operational framework coverin
On October 15, 2026, South Korean technology platform Kakao officially unveiled 'SPIRIT', a dedicated AI ethics framework, alongside 'NORMA', an operational governance system designed for autonomous AI agents, based on presentations delivered at its developer conference if(kakao)26 held on October 13 at the Yongin Kakao AI Campus.
Image source: Digital Today / Kakao
As artificial intelligence moves from conventional conversational interfaces toward autonomous agents capable of independently executing complex delegated tasks, Kakao's initiative establishes structural safeguards across the complete product lifecycle—from service planning and engineering design to deployment and continuous operations.
Six Core Ethical Principles for Autonomous AI Agents: 'SPIRIT'
To address the growing adoption of autonomous AI agents, Kakao has upgraded its existing artificial intelligence ethics guidelines into the 'SPIRIT' framework. The framework focuses on mitigating emerging risks during autonomous agent reasoning and establishing verifiable system trustworthiness across six core pillars:
- Safety: Ensuring that agent behaviors and outputs do not cause physical, psychological, or environmental harm to users or broader society.
- Privacy: Safeguarding personal data and sensitive user information across all processing stages while enforcing strict data minimization principles.
- Inclusion: Mitigating bias and discrimination to provide accessible, equitable, and fair services across diverse user groups.
- Responsibility: Guaranteeing explainability for agent actions and clearly delineating accountability across development and operational teams.
- User Agency: Ensuring that end-users preserve ultimate authority and direct intervention rights over agent tasks, even in highly autonomous workflows.
- Transparency and Trustworthiness: Openly detailing agent boundaries, operational mechanics, and data sources to foster accountable system trust.
Five Risk Management Dimensions and the 'NORMA' Implementation System
Moving beyond high-level ethical statements, Kakao introduced 'NORMA' as a concrete operational governance structure applicable to production engineering.
Kakao structured the potential risk surface introduced by autonomous agents into five core areas, directly pairing each with human oversight safeguards and emergency control mechanisms:
- Agency / Control: Preventing agents from exceeding delegated task scopes or making unauthorized unilateral decisions through structured human oversight and emergency override controls.
- Objectives: Reviewing assigned instructions and system goals prior to execution to verify alignment with ethical guidelines.
- Authority: Scoping and restricting delegated operational permissions and execution boundaries for agent tasks.
- Reference Resources: Auditing the integrity and factual reliability of external databases, documents, and toolkits leveraged during agent reasoning.
- Interaction: Preventing manipulative conversational tactics, deceptive outputs, or the transmission of harmful content to users.
To evaluate these safeguards during product development, Kakao outlined a three-stage AI Ethics Implementation Checklist:
- Stage 1: Service Characterization: Identifying domain-specific requirements, designated autonomy levels, and the types of data handled by the service.
- Stage 2: Proactive Risk Screening: Screening for potential misuse scenarios and safety vulnerabilities across the five risk dimensions.
- Stage 3: Control Measure Verification: Finalizing mandatory safeguards, including human oversight measures and emergency control mechanisms.
Child Protection via OECD-UNICEF 5Cs and Deployment Roadmap
The framework places particular emphasis on safety protections for children and adolescents. Kakao incorporated risk indicators derived from the OECD and UNICEF '5Cs' classification of online risks for children, tailoring these criteria specifically to autonomous AI agent interactions.
Kakao plans to begin phasing the AI Ethics Implementation Checklist into live production services before the end of 2026. However, detailed rollout timelines and organizational operational procedures remain subject to ongoing coordination across internal business divisions.
Furthermore, technical specifications outlining the exact runtime interruption protocols and system architecture enforcing emergency stops and human oversight have yet to be disclosed. Evaluating the practical enforcement efficacy of the framework will require monitoring further technical documentation as individual agent services roll out.