A Korea Advanced Institute of Science and Technology (KAIST) research team has been selected as the only institution in Korea for Microsoft's (MS) global artificial intelligence (AI) safety research program.
A team led by Professor Son Suel of the KAIST School of Computing said on the 28th that it was included as a recipient of research support for the External Red Team Alliance (EXTRA), operated by the MS AI Red Team.
EXTRA is a program that supports universities and research institutes in reviewing the security vulnerabilities and safety of AI systems from an external perspective. It was created to broadly address risks that are difficult to validate with only internal organizations at corporations, such as cybersecurity, multilingual environments, cultural context, AI alignment, and potential technology misuse.
More than 10 universities across six continents will participate in the program. In Korea, KAIST is the only one. Son's team will receive $25,000 (about 36.5 million won) in support as a gift award with no separate restriction on the research period.
The team is researching security and privacy technologies for Generative AI and large language models (LLMs). The main tasks are to analyze attack techniques targeting AI systems—such as model theft, personal data extraction, prompt injection, and membership inference—and to develop defensive technologies.
They also plan to study the security of AI agents used in real services, such as web agents and agentic browsers. The work addresses system-level defenses that reduce cases where AI performs unintended actions due to external inputs or malicious commands.
Professor Son Suel said, "As AI is rapidly applied to real services, securing not only performance but also safety and reliability has become important," and added, "We will analyze security threats that can arise in the field and develop response technologies."