KT said on Tuesday it is focusing on building an AI environment that can be used with confidence. [Photo: KT]

KT said on Tuesday it is focusing on building an AI environment that can be used with confidence by continuously strengthening a "responsible AI" framework premised on safety and trust, alongside advances in AI technology.

AI has recently been evolving into "agentic AI" that understands goals, makes its own judgments and carries out tasks. Across industries, expectations for technological innovation are rising, as are demands to ensure AI safety and reliability.

KT set AI ethics as a core value, beyond corporate compliance, to make a leap to AICT-centred management. In 2024, it established a dedicated organisation called the Responsible AI Center (RAIC) and appointed a CRAIO as the top officer, the first among South Korean telecom companies.

It enacted its own AI ethics principles, "ASTRI," and applies five core principles across the full process from AI planning, development and operations to use: accountability, sustainability, transparency, reliability and inclusivity.

KT also publishes an annual "Responsible AI Report" that summarises the company's various activities to strengthen AI ethics. In the report published in December last year, it detailed the social impact and safety evaluation process for its in-house AI models and its AI guardrail (SafetyGuard), contributing to the RAI open-source community. It also introduced practical cases of key activities based on a responsible AI framework, such as expanding RAI education.

KT continues to run a "Responsible AI Advisory Committee" made up of external experts by field. It actively participates in discussions on AI-related systems in cooperation with government and industry. It is also working to spread a responsible AI culture, including making AI ethics training mandatory for all employees, including those at affiliates and partner companies.

On the technology side, it reviewed domestic and overseas AI risk management frameworks and established its own AI safety standards suited to the South Korean environment. In October last year, it published a "Responsible AI Technical Report," which it described as a technical guideline for identifying, assessing and mitigating AI risks. The technical report establishes a lifecycle risk management framework from AI development to operations and presents an evaluation framework and implementation strategy applicable in real-world environments.

Separately, the AI guardrail that blocks harmful responses from AI models in real time was also released on the Hugging Face platform in September 2025. It recorded an F1 score of 97 under the "Kor Ethical QA" benchmark, which assesses the ethical response quality of Korean-language AI models.

As a result of these efforts, KT's in-house AI model "Mideum K 2.0 Base" became the first AI model in South Korea to obtain "AI Reliability Certification 2.0 (CAT 2.0)" from the Telecommunications Technology Association (TTA).

KT's responsible AI framework is also gaining attention on the global stage. The GSMA evaluates RAI maturity in four stages. In the 2025 assessment, KT achieved the top grade, "Advanced," described as embedded in corporate culture, and was introduced as a global best practice, the first among South Korean mobile carriers.

Bae Soon-min (배순민), a managing director and KT's CRAIO, said, "We will continue to contribute to the healthy growth of South Korea's AI industry and to securing user trust, based on AI technologies that can be trusted."

Keyword

#KT #GSMA #Hugging Face #TTA #Kor Ethical QA
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.