Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Responsible AI Technical Report

KT, :, Yunjin Park, Jungwon Yoon, Junhyung Moon, Myunggyo Oh, Wonhyuk Lee, Sujin Kim Youngchol Kim, Eunmi Kim, Hyoungjun Park, Eunyoung Shin, Wonyoung Lee, Somin Lee, Minwook Ju, Minsung Noh, Dongyoung Jeong, Jeongyeop Kim, Wanjin Park, Soonmin Bae

TL;DR

This work presents KT's Responsible AI framework, combining an AI Risk Taxonomy with a three-part RAI process (risk identification, assessment, and tooling) to operationalize responsible AI in domestic contexts. It couples qualitative and quantitative safety assessments with adversarial robustness testing (red-teaming) and delivers an integrated toolkit (Data Cleansing, Evaluation, Guardrail) to manage risks across data, development, and deployment. A key contribution is the SafetyGuard suite, including Prompt Guard and Content Guard, plus a multi-label guard for nuanced risk control, demonstrated against Korean benchmarks and real-time streaming constraints. The results show strong safety and robustness for Korean LLMs while highlighting gaps in socio-economic risk handling and the need for multimodal and domain-specific enhancements. Practically, the framework enables organizations to implement regulated, trustworthy AI through automated evaluation, real-time guardrails, and continuous taxonomy-driven improvements.

Abstract

KT developed a Responsible AI (RAI) assessment methodology and risk mitigation technologies to ensure the safety and reliability of AI services. By analyzing the Basic Act on AI implementation and global AI governance trends, we established a unique approach for regulatory compliance and systematically identify and manage all potential risk factors from AI development to operation. We present a reliable assessment methodology that systematically verifies model safety and robustness based on KT's AI risk taxonomy tailored to the domestic environment. We also provide practical tools for managing and mitigating identified AI risks. With the release of this report, we also release proprietary Guardrail : SafetyGuard that blocks harmful responses from AI models in real-time, supporting the enhancement of safety in the domestic AI development ecosystem. We also believe these research outcomes provide valuable insights for organizations seeking to develop Responsible AI.

Responsible AI Technical Report

TL;DR

This work presents KT's Responsible AI framework, combining an AI Risk Taxonomy with a three-part RAI process (risk identification, assessment, and tooling) to operationalize responsible AI in domestic contexts. It couples qualitative and quantitative safety assessments with adversarial robustness testing (red-teaming) and delivers an integrated toolkit (Data Cleansing, Evaluation, Guardrail) to manage risks across data, development, and deployment. A key contribution is the SafetyGuard suite, including Prompt Guard and Content Guard, plus a multi-label guard for nuanced risk control, demonstrated against Korean benchmarks and real-time streaming constraints. The results show strong safety and robustness for Korean LLMs while highlighting gaps in socio-economic risk handling and the need for multimodal and domain-specific enhancements. Practically, the framework enables organizations to implement regulated, trustworthy AI through automated evaluation, real-time guardrails, and continuous taxonomy-driven improvements.

Abstract

KT developed a Responsible AI (RAI) assessment methodology and risk mitigation technologies to ensure the safety and reliability of AI services. By analyzing the Basic Act on AI implementation and global AI governance trends, we established a unique approach for regulatory compliance and systematically identify and manage all potential risk factors from AI development to operation. We present a reliable assessment methodology that systematically verifies model safety and robustness based on KT's AI risk taxonomy tailored to the domestic environment. We also provide practical tools for managing and mitigating identified AI risks. With the release of this report, we also release proprietary Guardrail : SafetyGuard that blocks harmful responses from AI models in real-time, supporting the enhancement of safety in the domestic AI development ecosystem. We also believe these research outcomes provide valuable insights for organizations seeking to develop Responsible AI.

Paper Structure

This paper contains 34 sections, 8 figures, 9 tables.

Table of Contents

  1. Introduction
  2. AI Risk Identification
  3. AI Risk Taxonomy
  4. Severity Criteria
  5. RAI Assessment
  6. RAI Safety Assessment
  7. RAI Qualitative Assessment
  8. Harmlessness Assessment
  9. Helpfulness Assessment
  10. RAI Quantitative Assessment
  11. RAI Robustness Assessment
  12. Assessment Results
  13. RAI Qualitative Assessment Results
  14. Harmlessness
  15. Helpfulness
  16. ...and 19 more sections

Figures (8)

  • Figure 1: Overview of KT's Responsible AI
  • Figure 2: Structure of RAI Assessment
  • Figure 3: Ensuring Reliability of the RAI Qualitative Assessment
  • Figure 4: Overview of the red teaming dataset construction
  • Figure 5: RAI tools configuration across the AI lifecycle
  • ...and 3 more figures