The National Intelligence Service (NIS, Director Cho Tae -yong) announced on the tenth that it has discovered security problems by conducting technical verification on the Deep Chic, a generic artificial intelligence (AI) service, where domestic use is increasing rapidly.
Specifically, we identified that ▲ excessive personal information collection ▲ Use as a service learning data of all input data ▲ advertisers, user information sharing without restricted user information, and other advantages by language.
To start with, unlike the present AI service, he confirmed that he was collecting a ‘keyboard input pattern that would discover individuals’. As well as, he warned that chat records could possibly be transmitted by communicating with a Chinese company server (Volceapplog.com, etc.).
As well as, he identified that the usage of user input data as a learning data is lacking. The entire user’s information might be introduced and utilized into the training data.
Particularly, he added that the user’s service usage information is to be shared with advertisers. Because the holding period isn’t specified, it is feasible to share user information and unlimited storage without restrictions with advertisers.
Consequently, he said that it could cause a fatal problem of ‘storage of out of doors server’. Based on the Deep Chic Terms and Conditions, the people’s personal information and input data are stored within the Chinese server, and the Chinese government requests it if the Chinese government requests in line with the law.
Along with security issues, he identified the ‘imbalance’, which provides different answers by language. Within the case of other creating AI services similar to chat GPT and Clova X, the identical query has been answered no matter language.
Nevertheless, Deep Chic said that it’ll give different answers to the Northeast Fair, Kimchi, and Dano. Particularly, within the query of Kimchi’s origin, Korean is alleged to be ‘representative foods with Korean culture and history’, but in China, ‘origin isn’t Korea, not Korea.’
Meanwhile, the NIS distributed a letter to government departments that emphasized security attention when using AI work similar to ‘Deep Chic’ on the third. In the longer term, the NIS plans to conduct a detailed inspection on the technology safety of ‘Deep Chic’ in cooperation with related organizations. If obligatory, he’ll further explain the outcomes of the inspection to the people.
By Jang Se -min, reporter semim99@aitimes.com
