Samsung SDS outlines AI-driven software self-healing at KIISC workshop
A Samsung SDS master presented research on AI vulnerability detection and patch automation at the 2026 AI x RMF Workshop in Seoul on October 22.
Samsung SDS master Jo Ji-hoon presented a session titled "SW Self-Healing at AI Scale" at the KIISC "2026 AI x RMF Workshop," held at Samsung SDS West Campus in Seoul on October 22, according to dailysecu.com. The talk covered AI-based software vulnerability detection and patch automation, along with related Samsung SDS research cases.
According to the presentation materials, teams in the 2025 AIxCC finals performed automated vulnerability analysis and patching against about 54 million lines of real source code, finding 54 of 63 synthetic vulnerabilities and successfully patching 43.
Samsung SDS is pursuing automation of source code vulnerability checks, penetration testing and design verification based on AI agents in cooperation with OpenAI.
Test generation and cryptography research
The company has researched EvoFuzz, which automatically generates Java unit test code, and recently expanded the work to EvoLLM, which combines a large language model. In comparative experiments cited in the presentation materials, EvoLLM recorded 71.9% code line coverage, 57.1% branch coverage and a mutation testing score of 49.6%.
Samsung SDS is also researching S-CAPE (Samsung SDS Crypto Agility Platform for Enterprise), which automatically identifies cryptographic usage.
Quick answers
What did Samsung SDS present at the KIISC 2026 AI x RMF Workshop?
Jo Ji-hoon gave a talk titled "SW Self-Healing at AI Scale" on October 22, covering AI-based vulnerability detection and patch automation and related Samsung SDS research.
What results did EvoLLM record in the cited experiments?
71.9% code line coverage, 57.1% branch coverage and a mutation testing score of 49.6%.