Chinese J-36 engineers warn AI can fabricate radar, weapon, and combat data

August 23, 2026 at 10:18 AM2 min

This digest was compiled by AI from multiple sources — links to the originals are below.

Engineers at AVIC Chengdu Aircraft Research and Design Institute, the organization behind China's J-20 and next-generation J-36 fighter programs, warned that large language models can generate highly convincing but false radar, weapon, and combat data. The warning appeared in a June 20 article in the journal Information Studies: Theory & Application, operated by state-owned defense firm Norinco. The fabricated data could distort fighter design decisions and military intelligence assessments.

Key Facts

The warning was published on June 20 in Information Studies: Theory & Application, a journal operated by China's state-owned defense company Norinco.
Engineer Zhang Xianzhe, affiliated with the Chengdu aircraft design organization, stated that large language models can produce extremely convincing but non-existent information.
AI-generated errors can include fabricated fighter dimensions, payload capacity, weapon capabilities, maximum speed, combat radius, radar scan distances, and frequency bands.
Zhang recommended restricting models to reliable military data, creating searchable knowledge bases, using clearer commands, and cross-checking outputs with different AI systems.

The Warning

Engineers from AVIC Chengdu Aircraft Research and Design Institute, the organization behind the J-20 fighter and the next-generation J-36 program, issued the warning. The assessment appeared in a June 20 article in Information Studies: Theory & Application, a journal operated by the state-owned defense firm Norinco. Engineer Zhang Xianzhe, linked to the Chengdu aircraft design organization, said large language models can generate highly convincing but non-existent information.

Design and Intelligence Risks

During fighter development, designers must analyze enemy radar coverage, operating frequencies, missile engagement zones, electronic warfare capabilities, early warning aircraft positions, and ground-based air defenses. These data shape fundamental design decisions such as stealth characteristics, sensors, maneuverability, and combat range. Zhang said AI can fabricate a fighter's dimensions, payload capacity, weapon capabilities, maximum speed, or combat radius, as well as radar scan distances and frequency bands. Because AI output appears technically consistent, analysts or engineers may treat false information as genuine intelligence data. Zhang also warned that AI could recommend materials without adequate durability, design aerodynamically impossible maneuvers, or prepare mission plans exceeding the aircraft's operational range.

Mitigation Measures

In military intelligence, AI can create non-existent bases, units, and troop movements, which, if fed into simulations, could produce completely erroneous assessments of an adversary's military capacity. Human analysts making final decisions can reduce risk, but as the volume of AI-generated information grows, detailed verification of each output becomes increasingly difficult. Zhang recommended that defense organizations restrict models to reliable military data and create searchable knowledge bases where information can be verified. Using clearer and more precise commands can reduce the likelihood of ambiguous interpretations by the model. Zhang also proposed cross-checking AI conclusions with different AI systems that question errors in each other's outputs.