[Edaily Reporter Lee So-Hyun ] Naver (NAVER(035420)) is expanding the scope of its artificial intelligence (AI) safety management from a focus on technical models to include actual services and user experience.

Naver announced on the 8th that it has unveiled “ASF 2.0,” an AI safety framework that expands its scope to encompass users’ safe AI service experiences.

ASF 2.0 is an updated version of the ASF AI safety framework that Naver unveiled at the 2024 AI Seoul Summit. While the original ASF focused on managing the performance and risk levels of AI technical models, ASF 2.0 centers on managing safety based on the services users actually experience.

At the AI Safety Seoul Forum held that day in Gangnam-gu, Seoul, Naver shared the direction and components of ASF 2.0, its enhanced classification, management, and enforcement framework, and its future action plan.

Song Dae-seop, Naver’s AI Safety Policy Leader, who delivered the presentation, explained, “As the technological, service, and policy/regulatory environments surrounding AI evolve, the challenge has expanded beyond simply making a single model safe to the question of how to safely design and operate services used by tens of millions of people by combining various AI models.”

ASF 2.0 reflects Naver’s “On-Service AI” strategy, the growing prevalence of multi-model environments in the global AI ecosystem, and changes in the policy and regulatory landscape, such as the enactment of the Framework Act on Artificial Intelligence. On-Service AI is a strategy in which Naver integrates AI into its existing services—including search, shopping, local services, and content—to enhance the user experience.

ASF 2.0 aims to manage safety at every stage, from the launch to the overall operation of AI services. To this end, the scope of management—which was previously centered on individual technical models—has been expanded to cover AI services operating in a multi-model environment.

The evaluation criteria have also been refined from a single, performance-centric standard to include context, use cases, and impact. Specifically, the AI risk classification system categorizes potential risks that may arise from AI services. Subsequently, the expected impact is assessed based on the AI impact assessment matrix, taking into account the service’s application area and scope. Finally, based on continuous safety evaluations and user feedback, the system ensures that users can safely utilize AI services.

Naver has also established a company-wide implementation framework called “CHEC 2.0” to ensure that ASF 2.0 is consistently applied throughout the service launch process. CHEC 2.0 is a consultation framework designed to review the ethics and safety of human-centered AI.

CHEC 2.0 was also applied to the “AI Tab,” which was unveiled last June. Naver explained that it reviewed AI safety throughout every stage, from the design of the AI Tab to its launch.

Naver plans to continuously assess the safety of AI-based services—including the AI Tab—that are scheduled for release or have already been launched, using CHEC 2.0.

Song, the team leader, emphasized, “By unveiling the Naver AI Ethics Guidelines in 2021 and NAVER ASF Beta in 2024, Naver has been continuously refining its AI safety framework in line with global trends in AI ethics and the domestic policy environment.” He added, “We will continue to accumulate experience and know-how regarding AI safety and share and advance this knowledge through collaboration with academia, policymakers, external experts, and various institutions.”