Naver on Wednesday unveiled "ASF (AI Safety Framework) 2.0," an AI safety framework that expands its management scope to cover users' safe artificial intelligence (AI) service experience.
ASF 2.0 is an AI safety management framework that updates the ASF that Naver unveiled at the 2024 AI Seoul Summit, incorporating an expanded concept that extends to user protection. The key is that it broadens the safety management scope of the existing ASF, which focused on managing the performance and risk levels of AI technology models, to manage safety centered on the services that users actually experience.
Naver shared the direction and components of ASF 2.0, along with its strengthened classification, management and execution systems and future activity plans, at the Seoul Forum on AI Safety & Security (SFASS), held in Gangnam-gu, Seoul, that day.
Song Dae-seob, Naver's AI Safety Policy Leader, who gave the presentation, explained the background behind establishing ASF 2.0. "As the technology, services, policies and institutional environment surrounding AI change, the issue is expanding beyond making a single model safe to how to safely design and operate services used by tens of millions of people by combining various AI models," he said.
Song explained that Naver's ASF 2.0 reflects multiple factors, including Naver's advancing "On-service AI" strategy such as the AI Tab and shopping AI agents; the spread of multi-model environments in the global AI ecosystem; and changes in the policy and institutional environment such as the enactment of the AI Basic Act.
ASF 2.0 aims to safely manage AI from the perspective of users and services, managing safety at all stages from the launch of AI services through their overall operation. To this end, it expands the management scope, which was previously centered on AI technology models, to AI services based on multi-model environments, and also subdivides the single performance-focused evaluation criterion into context, use-case and impact.
Specifically, it categorizes risks that can arise in AI services according to the AI Risk Taxonomy, and evaluates the expected impact according to the area and scope of use in line with the AI Impact Assessment Matrix. Finally, based on this system, it manages AI services through continuous safety evaluation and user feedback so that users can use AI services continuously and safely.
Naver has also established "CHEC 2.0," a company-wide execution system that helps ensure ASF 2.0 is consistently applied during the service launch process. The "AI Tab," introduced this June, also underwent AI safety checks throughout all stages from design to launch via CHEC 2.0. Naver plans to continuously check the safety of AI-based services, including the AI Tab, that are scheduled to launch or have already launched, through CHEC 2.0.
"By unveiling the 'Naver AI Ethics Principles' in 2021 and 'NAVER ASF Beta' in 2024, Naver is continuously refining its AI safety framework in line with the global trend in AI ethics and the domestic policy environment," Song said. "Going forward, we will accumulate experience and know-how on AI safety, and share and develop this experience and know-how through collaboration with academia, policymakers, external experts and various institutions."
