DuckDuckGo, a search engine that positions itself as privacyfocused, offers Duck.ai—an AI chat service that allows users to interact with models such as Anthropic’s Claude, Meta’s Llama, and OpenAI’s GPT-4/5, while anonymizing user data. According to DuckDuckGo, Duck.ai does not track user behavior or store conversation content (conversations are reportedly stored on the user’s device rather than on remote servers), and user data is not used for AI training. All metadata containing personal information— such as IP addresses—is completely removed before messages are sent to model providers like Anthropic or OpenAI. In other words, these providers cannot identify who sent a given message. However, even when using Duck.ai, this does not mean that personal data or confidential information included in prompts is fully protected. Information contained in prompts is still transmitted to AI providers via Duck.ai. That said, DuckDuckGo states that it has contractual agreements with AI providers requiring them to delete all received data once it is no longer necessary to generate a response (within a maximum of 30 days, subject to limited exceptions for safety and legal compliance). Although Duck.ai currently has functional limitations compared to other commercial generative AI services when used in Korea, it offers relatively strong security protections. Depending on the intended use case, organizations may therefore wish to consider this service as an option. Fifth, if there are concerns about the security of data shared through generative AI services, it is necessary to regularly back 82 83