China wants AI to run nuclear reactors – but can it be trusted?

China aims to revolutionize the nuclear sector with AI

At this year's World Artificial Intelligence Conference (WAIC) in Shanghai, researchers from the Chinese Academy of Sciences (CAS) put forward one of the most controversial technology proposals the energy sector has seen in a long time: full integration of artificial intelligence into the operation of nuclear power plants.

The initiative has been named "AI for ADANES" – where ADANES stands for Accelerator-Driven Advanced Nuclear Energy System. According to CAS, this is not merely an upgrade of existing technology, but a fundamental shift in the safety logic governing conventional reactors, as reported by OilPrice.com.

What is ADANES – and why is it different?

ADANES is a subcritical nuclear energy system that combines three functions in a single platform: breeding of nuclear fuel, transmutation of spent fuel, and electricity generation. The subcritical operating mode is the technology's most important safety feature – the system cannot sustain an independent chain reaction and is therefore structurally harder to drive out of control.

The AI architecture is built around a so-called "triple-driving" principle, combining raw data, physics-based models, and expert knowledge across five architectural layers.

The project's engineering platform, CiADS (China Accelerator Driven Sub-critical System), is currently under construction and will serve as the technology verification facility for the entire ADANES concept.

The "black box" meets zero tolerance

The most contentious question surrounding the project is not reactor physics, but the AI itself. Complex neural networks are notoriously difficult to understand from the inside – they deliver results without human operators always being able to trace the decision-making process. The nuclear industry does not permit this kind of ambiguity.

He Yuan, assistant director of the Institute of Modern Physics and technical director of ADANES, openly acknowledges the tension. According to him, the project's approach is designed to reconcile precisely "the opacity and unpredictability of black-box AI systems with the stringent safety requirements of the nuclear sector," as reported by OilPrice.com.

The core contradiction is real: AI is not designed for zero tolerance, but nuclear power demands exactly that.

To address this challenge, Antuo – a key player in the project – has launched a so-called "Unified Data Source Platform." The platform consolidates data from the entire CiADS facility lifecycle – from requirements and simulations to operational data – into a single unified system that the AI can interpret and from which digital twins can be developed. The goal is to make AI decisions traceable and verifiable.

International frameworks set strict requirements

There are no known public statements from international nuclear authorities or other countries specifically commenting on the ADANES project as of July 2026. That does not mean the broader debate is absent.

The EU's AI Act, which entered into force in 2024, classifies AI systems in critical infrastructure as high-risk and imposes requirements for transparency, human oversight, and demonstrable reliability. The US National Institute of Standards and Technology (NIST) published its AI Risk Management Framework in 2023, closely tied to the very principles of traceability and accountability in complex systems.

The OECD's AI Principles from 2019 – adopted by more than 40 countries – emphasize that AI in critical applications requires transparent decision-making and that human oversight must be meaningful, not merely symbolic.

What does this mean for energy markets?

If the ADANES concept is validated through the CiADS platform, it could represent a technological breakthrough for nuclear waste management and fuel cycles – two of the largest cost drivers in the global nuclear power industry. A successful AI integration could also accelerate China's ambitions to become a global exporter of next-generation nuclear power technology.

For now, the project is in an early validation phase, and the requirements of international AI ethics frameworks – particularly the demand for explainability – will represent one of the most demanding tests of the system's credibility outside China's borders.

Sources: OilPrice.com, CAS/WAIC presentations, OECD AI Principles, EU AI Act, NIST AI RMF