Gaebowi Recommends Information Processing Improvements to AI Operators Including SKT AIDOT and Naver Snow

Status Inspection of Four AI Application Service Providers

The Personal Information Protection Commission recommended improvements in information processing to businesses providing artificial intelligence (AI) application services, such as SK Telecom and Snow, a subsidiary of Naver.

On the 12th, the Commission held a plenary meeting to review and resolve the results of a preliminary inspection of SK Telecom's A. (A-dot), Snow, DeepL, and VUNO. This inspection aimed to proactively identify vulnerabilities in personal information processing amid the rapid expansion of AI services and to encourage improvements.

First, SK Telecom's call summary and real-time interpretation service A. did not leave access logs in the system that stores call text files. Since access logs must be stored and reviewed according to system requirements, the Commission recommended corrective measures. Additionally, it advised minimizing the retention period of text files and strengthening de-identification in accordance with personal information protection principles.

Snow, which provides an AI profile image generation service, had shortcomings in sufficiently reviewing the safety of external development tools (SDKs). Therefore, the Commission recommended that when using external SDKs, companies should check whether there is a possibility of processing or transmitting unintended personal information.

DeepL, a translation service, conducted AI training and human review on user-input information but did not clearly disclose this. However, during the inspection process, this was corrected, so no separate recommendation was made. In the case of VUNO, an AI medical imaging interpretation service, no violations related to data collection and processing laws were found.

A Commission official stated, "We will continue to monitor AI application services" and added, "We plan to establish personal information protection measures and safe personal information utilization plans for the AI era."