KAIST, the Only Domestic Institution Selected for Microsoft's AI Safety Research Program

As generative AI rapidly expands across industries and society, technologies ensuring AI safety and security are emerging as new core competitive advantages. KAIST has been selected as the sole South Korean institution to join Microsoft's global AI safety research program, advancing the development of next-generation AI security and reliability technologies. The university's Professor Soon-Sool was chosen as a research support recipient for Microsoft's EXTRA program, a new global initiative by the Microsoft AI Red Team to verify AI system safety and security vulnerabilities from diverse perspectives. This program aims to expand an independent AI safety research ecosystem by involving universities and research institutions worldwide, moving beyond internal corporate evaluations. Microsoft established EXTRA to comprehensively assess AI safety across cybersecurity, multilingual environments, cultural contexts, AI alignment, and misuse risks, which cannot be fully evaluated by a single organization. KAIST will receive a $25,000 research grant for AI safety, security, alignment, and responsible AI development, with the funds supporting its research team's efforts. EXTRA now includes over 10 universities across six continents, with KAIST as the sole South Korean participant. Microsoft plans to strengthen academic-industry collaboration through this program. Professor Soon-Sool's team focuses on security and privacy protection technologies for generative AI and large language model-based systems, analyzing adversarial attack techniques targeting deep neural networks and language models.