appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Zhang Tong

Large language models are expert in intelligence work, able to quickly mine key data and analyse weapons performance – tasks that once took analysts huge amounts of time and effort.
But engineers at China’s AVIC Chengdu Aircraft Research and Design Institute, developer of the People’s Liberation Army’s backbone J-20 and next generation J-36 stealth fighters, have urged caution.
In a paper published on June 20, engineer Zhang Xianzhe said that “AI models can invent aircraft specifications – length, payload, weapons, speed, combat radius – and get radar scan ranges and frequency bands wrong”.
“The so-called hallucination effect, where models produce plausible but false outputs, could have severe consequences in the high-stakes, low-error world of defence intelligence, possibly even causing strategic miscalculations,” he said.
Zhang’s work was published in Information Studies: Theory & Application, a journal run by China’s state-owned arms maker Norinco.




According to the paper, every fighter jet begins with a deep understanding of the enemy and the battlefield. Designers must answer critical questions: what is the opponent’s radar coverage and which frequencies do they use?
They must also consider how large the no-escape zone of their air-to-air missiles is, which bands their electronic warfare systems can cover, where their early-warning aircraft patrol and how their ground-based air defence network is laid out.
These factors determine stealth design, radar power, electronic warfare architecture, manoeuvrability, range and combat radius. In short, intelligence gathering was the very first step in building a fighter, Zhang wrote.
But he warned that “since large language models are rarely trained on authoritative military sources, they often generate content that defies basic physics, design principles, or operational constraints”.
This could include “specifying materials that exceed fatigue limits, proposing flight manoeuvres that violate aerodynamics, or drawing up mission plans beyond an aircraft’s actual range”.
Zhang said that when it came to defence technology, AI could generate answers that looked reasonable but were factually incorrect – not just aircraft parameters, but also radar scan ranges, frequency bands, and even entirely fictitious military bases, units, or exercise movements.
“The most direct harm comes in intelligence analysis and simulation, where acting on such hallucinations can lead to false assessments of combat capability,” he added.


Real-world incidents have already occurred. In March 2026, a US strike on Iran hit a school, killing more than 175 children. Outdated data – the building was still marked as a military target – and an AI system that flagged it as high-priority were blamed.
According to officials, humans made the final decision but the sheer speed and volume of AI‑processed intelligence raise serious doubts about whether frontline analysts can thoroughly verify every AI-recommended target in time.
In his paper, Zhang suggested several ways to cut AI hallucinations: feed models with trustworthy defence data, build searchable knowledge bases for reference, write clear and specific prompts, and use cross-checking through AI debate.
He also called for the promotion of safer, more efficient, and more reliable applications of large language models in defence technology intelligence.
Passionate about science? Dive deeper with the Dark Matters newsletter, a weekly in-depth analysis on China’s rise in science, technology and military that goes beneath the surface. Sign up for free now.


T. rex was warm-blooded and had body temperature close to ours, study finds

China uses 100,000 home-grown AI chips to build leading weather forecast system

Geologists find unexpectedly large water sources in Xinjiang beneath China’s largest desert

Chinese scientists build world’s smallest CT scanner, weighing just 6kg

Science

China’s J-36 fighter jet designers report danger of military AI hallucinations

China’s next-generation stealth fighter is unofficially known as the Chengdu J-36. Photo: X / Rupprecht_A

Chinese missile AI tracks F-22, F-35-like heat signatures with over 90% accuracy

The data set included three categories of airborne targets, including two types of aircraft mimicking the F-22 and F-35. Photo: US Department of Defence

China adds an 800-passenger jet with ‘flying wing’ body to its big aircraft plan

The unnamed test model, which would be capable of carrying 800 passengers, is pictured in a wing tunnel. Photo: Journal of Aerospace

Chinese drone maker DJI gets another shot at Pentagon blacklist removal

DJI products like the Mavic 2 Pro have proven popular in the US. The Shenzhen-based company is challenging its inclusion on a Pentagon blacklist of firms allegedly linked to the Chinese military. Photo: Getty Images

A DJI Matrice 350 RTK, operated by New Jersey Transit, flies over an emergency response drill and training exercise in East Rutherford, New Jersey, on April 18. Photo: Reuters

News

China

Content generated by artificial intelligence models can ‘defy basic physics, design principles or operational constraints’


Content generated by artificial intelligence models can ‘defy basic physics, design principles or operational constraints’.


China’s J-36 fighter jet designers report danger of military AI hallucinations

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Content generated by artificial intelligence models can ‘defy basic physics, design principles or operational constraints’

Content generated by artificial intelligence models can ‘defy basic physics, design principles or operational constraints’