Nuclear reactor designers must introduce physical artificial intelligence through low-level auxiliary tasks like laboratory automation and document analysis before expanding into core operations, according to experts speaking on November 12 at the DETO 2026 technology forum in Hannam-dong, Seoul.
Reactors leave no room for minor operational errors, making safety validation the primary hurdle for advanced software integration.
Seo said that physical AI deployment requires answering structural questions regarding autonomy levels and verification methods rather than merely selecting model architectures.
“The nuclear industry requires higher reliability and safety than other fields,” Seo told edata.ekn.kr, adding that deployment must begin with routine tasks like document discovery, knowledge organization, and interpretation before handling complex design phases.
Seo advocated for a supervised autonomy approach that restricts AI agents within isolated systems. Under this framework, models handling sensitive tasks operate on internal local servers without external connections, while unpredictable outputs trigger mandatory human intervention.
Corporate leaders deploying AI for primary equipment design and facility inspections point to regulatory uncertainty as a major roadblock. Se-young Chang, managing director of AI Innovation at Doosan Enerbility, explained that nuclear components must adhere to strict international codes.
“Because primary nuclear equipment must be designed and manufactured in accordance with international standards and codes, a major homework assignment is how to obtain international certification for AI utilization,” Chang told edata.ekn.kr. He added that while phased array ultrasonic testing handled by AI outperforms human operators, the lack of global certification limits commercial deployment.
Chang said that Doosan plans to participate in a new project next year to incorporate AI inspection processes into the AZTEC code. Ki-su Kim, CEO of DeepEye, added that regulatory bodies should clarify AI rules gradually to encourage broader industry adoption.
