KDDI and KDDI Research, Inc. are conducting research and development on AI security based on the “KDDI Group AI Development and Utilization Principles.” The AI Security Portal organizes, classifies, and centralizes the latest information on the web about attacks and defense methods against AI itself, as well as the impact on people and society and the technologies that can be used to counter them. The site comprehensively disseminates information on AI security, including an overall picture of AI security, individual commentary articles, literature information, news and event information, and related sites. Users can use this site as a starting point to obtain the latest information on AI security. In order to consolidate and provide information on AI security, KDDI Research, Inc. has built a knowledge aggregation environment that can efficiently collect and organize information on AI security on the web by systematizing knowledge and technology.
In addition to attacks and defense methods against AI itself (information system aspects), we also conducted research on the elements of the negative impact that AI has on society and people as a result of attacks and misuse of AI, such as privacy violations and copyright infringements (external action aspects), whose relationship with AI security has not been systematically organized until now. Based on the results, we created an AI security map that organizes the relevance of AI security information. Since the negative impact of attacks and misuse of AI is systematically organized, it is possible to know the extent of the impact that attacks on AI have on people and society, which contributes to the consideration of countermeasures.
In addition, we have developed and introduced a new classification technology using LLM to classify the vast amount of literature information on AI security on the web. This technology automatically assigns appropriate labels to literature and classifies it in order to organize literature such as papers published daily on the web in an easy-to-understand manner. This makes it possible to organize and classify the latest AI security information that has not been organized before, and newly published literature is also accumulated in the database from time to time, making it easy for users to access the latest information they are looking for.
Also Read: Araya launches “YORO-ZOO,” a service that delivers pre-configured edge AI
KDDI and KDDI Research will continue to provide information and warnings on AI security, and promote awareness of risk countermeasures to the entire society. The website is one of the results of research and development carried out under the research and development theme “Systematization of knowledge and technology for safe and secure AI utilization and construction of a knowledge aggregation environment” in the “Establishment of AI security technology necessary for a data-driven economic society where artificial intelligence (AI) permeates” program of the Economic Security Key Technology Development Program (K Program) promoted by the National Research and Development Agency, Japan Science and Technology Agency. This result will enable the comprehensive and rapid systematization of the latest AI security technology and knowledge, and the establishment of optimal risk communication methods through highly accurate automatic collection and classification, which will contribute to the realization of the research and development initiative’s goal of accumulating necessary knowledge about AI security and organizing and acquiring knowledge and technology systems.
SOURCE: Yahoo
