Naver has disclosed a safety management system that breaks down risks that can arise in AI services into 110 detailed items and applies it from service design through operations.

Naver said on Tuesday it published for the first time an "AI Safety Progress Report" that includes the real-service application process and execution cases of its in-house AI safety system, ASF 2.0.

ASF 2.0 is Naver's AI safety management system that includes not only the performance and risk level of AI models, but also protection of users and services. It focuses on managing safety across the entire process from service design and development to launch and operations.

The report includes how ASF 2.0 and the execution system based on it, CHEC 2.0, are applied to real services.

Based on legal, social and industry cases, Naver built N-ARTI, which classifies risks that can arise in AI services into 110 detailed items. It is a system to identify and classify AI risks by applying the same standards across the company while reflecting the characteristics of each service.

It can also be used to build guardrails to prevent generative AI from producing answers that deviate from pre-set safety standards.

It also established N-ASET to assess the safety of AI answers. The tool evaluates AI responses from two perspectives, "harmfulness" and "usability". It checks not only whether responses contain harmful content, but also whether they appropriately answer users' questions and include necessary safety elements such as disclaimers and guidance to experts.

It also disclosed real-service application cases. Naver is indicating and notifying users in its AI shopping agent and weather service that AI is being used. It also applied CHEC 2.0 to the "AI tab" launched in June to check safety from design through launch.

Naver plans to expand cooperation with academia and specialised institutions, including the Korea Advanced Institute of Science and Technology (KAIST), Seoul National University's AI Policy Initiative (SAPI), Korea University's AI Safety Research Center and the AI Safety Institute (AISI), to continue research to identify and prevent and mitigate risks in AI services.

Won Seong-jae (원성재), head of Naver's AI Safety Center, said, "As AI technology rapidly permeates users' daily lives, AI safety is also entering a stage where it must be addressed concretely in real service environments beyond theoretical discussions." He said, "We will continue to advance and apply ASF 2.0 so that human-centered values are placed as the top priority throughout the entire process of AI development and use, and so that risks that can arise in services can be prevented and mitigated."