NAVER Publishes Examples of Applying Its ‘AI Safety Framework’

[Financial News] NAVER has published its first AI Safety Progress Report, detailing how its in-house artificial intelligence (AI) safety framework, ASF 2.0, is applied to actual services and providing specific examples of its implementation. The report aims to establish NAVER’s own AI safety management framework by covering AI model performance and risk levels, as well as the protection of users and services.
According to industry sources on the 16th, the report includes specific examples of how ASF 2.0 and the execution framework based on it, CHEC 2.0, are applied and operate in actual services. ASF 2.0 is designed to systematically manage safety throughout the entire lifecycle of AI services—from design and development to launch and operation—while reflecting NAVER’s On-Service AI strategy and changes in the global AI ecosystem, policies, and regulatory environment.
NAVER first established N-ARTI, an AI risk classification and identification framework that categorizes potential risks in AI services into 110 detailed items based on legal, social, and industrial cases. N-ARTI systematically identifies and classifies AI risks while reflecting the characteristics of each service and applying consistent safety standards across the company. Based on this framework, NAVER can establish guardrails to ensure that information provided by generative AI services does not exceed predefined safety standards.
NAVER also developed N-ASET, a tool for evaluating the safety of responses provided by generative AI services. N-ASET comprehensively assesses AI service responses from two perspectives: harmfulness and usability. Rather than simply blocking harmful responses, it also examines whether the service appropriately addressed users’ queries, enabling a balanced evaluation of safety and service usefulness.
The report also introduced cases in which NAVER displayed and disclosed the use of AI in services such as its AI shopping agent and weather service to inform users. The initiative is intended to communicate proactively so users can intuitively understand how AI safety measures are implemented in actual services. AI Tab, introduced last June, is another representative example: its AI safety was reviewed throughout the entire process, from design to launch, based on CHEC 2.0.
NAVER plans to continue its research and practical efforts to proactively identify, prevent, and mitigate risks that may arise in AI services. It will also expand cooperation with academic and professional institutions in Korea and abroad, including the Korea Advanced Institute of Science and Technology (KAIST), the Seoul National University AI Policy Initiative (SAPI), the Korea University AI Safety Research Center, and the Artificial Intelligence Safety Institute (AISI).
Won Seong-jae, head of the NAVER AI Safety Center, said, "As AI technology rapidly becomes embedded in users’ daily lives, AI safety has entered a stage in which it must be addressed concretely in actual service environments, beyond theoretical discussions. NAVER will continue to advance and apply ASF 2.0, placing human-centered values first throughout the entire process of AI development and use and working to prevent and mitigate risks that may arise in AI services."
[email protected] Jang Min-kwon Reporter