0
Cover of Naver AI Safety Progress Report. / Naver

Naver has established a comprehensive safety evaluation framework that categorizes potential risks in AI services into 110 granular criteria, vetting safety from initial design through launch and ongoing operations.

On the 16th, Naver announced the inaugural publication of its AI Safety Progress Report, detailing real-world implementation processes and case studies under its proprietary AI safety framework, ASF 2.0. Upgraded to oversee user and service safeguards in addition to AI model performance and hazard levels, ASF 2.0 implements its operational framework, CHEC 2.0, across live services.

Naver established N-ARTI, which categorizes potential legal, social, and industrial AI risks into 110 specific items. Designed to reflect individual service characteristics while identifying risks against unified company-wide standards, N-ARTI serves to construct guardrails ensuring generative AI outputs remain within predefined safety boundaries.

The company also introduced N-ASET to evaluate AI-generated responses. This system assesses both harmfulness—examining whether an answer could cause harm to users—and usability, evaluating whether the AI responded appropriately while providing necessary disclaimers or guidance toward human professionals.

These safety standards have already been applied to live offerings. Launched in June, the AI Tab underwent safety assessments from design to rollout based on CHEC 2.0, while the AI Shopping Agent and weather services explicitly inform users that AI is being utilized.

Won Sung-jae, Head of Naver AI Safety Center, stated, "AI safety is entering a phase where it must be addressed concretely within actual service environments, moving beyond theoretical discussions," adding, "We will continue to advance and apply ASF 2.0 to prevent and mitigate potential risks that may arise across AI services."

                                                                                                          Lee Seo-yeon