KAIST said a research team led by computer science professor Minsoo Kim developed a reinforcement-learning technique, RL-SPH, that trains AI to produce plans that satisfy real-world constraints without external optimisation software. The system searches first for a feasible plan and then reduces cost and time while keeping constraints. In benchmark tests it found feasible plans in all cases, improved solution-quality metrics and reduced search and training time versus other AI methods.

Korea Advanced Institute of Science and Technology said it developed a safety verification technology that can uncover about seven times more diverse hidden vulnerabilities in generative AI than existing methods. The team built a new red-teaming framework, Stable-GFlowNet, to verify large language model safety. Tests found 134 unique attack types versus 17 previously, while maintaining a 92 percent attack success rate. The work was selected as an ICML 2026 spotlight paper.