Reporting date: November 3, 2026
Scope: Chinese military doctrine, government research, intelligence organizations, commercial AI systems, procurement programs, and documented information operations.
Source base: Research, translated Chinese publications, uploaded documents, and investigative findings accumulated in the PRC project.

Executive assessment

China is developing the doctrinal, technical, and organizational foundations for LLM-enabled cognitive warfare. The evidence points toward an emerging model of information operations that combines population surveillance, behavioral profiling, synthetic-content generation, autonomous agents, and continuous assessment of audience reactions. This differs from traditional propaganda. The proposed architecture does not simply distribute a centrally approved message to a large audience. It seeks to understand individual and collective cognitive characteristics, tailor information to those characteristics, influence behavior through sustained interaction, and modify subsequent messaging according to observed results.

In People’s Liberation Army (PLA) military doctrine, cognitive warfare, or cognitive-domain operations (认知域作战), seeks to influence an adversary’s perceptions, beliefs, emotions, judgment, and decision-making to achieve strategic and military objectives. Chinese military theorists regard the cognitive domain as a battle-space alongside land, sea, air, space, cyberspace, and the electromagnetic spectrum, where controlling information and shaping how populations interpret events can weaken an adversary’s political cohesion, public confidence, and willingness to fight.

Cognitive warfare extends beyond traditional propaganda and psychological operations by exploiting social media, information algorithms, artificial intelligence, behavioral profiling, and increasingly large language models (LLMs) to identify audiences, manipulate information exposure, construct persuasive narratives, and induce desired reactions. PLA publications describe integrating these activities with conventional military operations while employing civilian institutions, commercial technology platforms, and information networks to extend their reach.

The intended effect is not necessarily to convince an adversary of a particular falsehood, but to shape the information environment and cognitive processes through which individuals and societies understand events, make decisions, and respond to conflict. In this framework, cognitive warfare aims to degrade an adversary’s ability to make independent, coherent decisions while advancing Chinese political and military objectives, potentially achieving strategic effects before conventional hostilities begin.

The project’s evidence establishes three distinct developments.

• First, the People’s Liberation Army has articulated a theory of algorithmic cognitive warfare. Chinese military publications describe AI and algorithms as mechanisms for influencing perception, exploiting psychological vulnerabilities, and shaping political decisions. The 2023 paper Mechanisms of Effectiveness and Key Technologies of Algorithmic Cognitive Warfare provides one of the clearest descriptions of this approach.

• Second, Chinese researchers are extending that theory into LLM-driven cognitive agents and multi-agent simulation. Publications from 2024 through 2026 describe systems capable of modeling social behavior, generating personalized narratives, maintaining persistent digital identities, coordinating multiple agents, and evaluating influence strategies through simulation.

• Third, commercial and government-linked capabilities increasingly overlap with the proposed military architecture. Chinese firms have developed or advertised social-media monitoring, behavioral analysis, information recommendation, synthetic-content generation, and cognitive decision-support systems. Separately, public threat-intelligence reporting has documented Chinese-linked influence operations using generative AI.

The most important intelligence distinction is that the individual capabilities are documented, but their integration into a single, deployed PLA or Ministry of State Security cognitive-warfare platform remains unverified.

Key judgments

Judgment	Confidence
PLA-affiliated researchers advocate algorithmic and AI-enabled cognitive warfare.	High
Chinese military and academic researchers are investigating LLM agents for cognitive operations and social simulation.	High
Chinese commercial firms possess technologies applicable to automated influence operations.	High
Chinese-linked influence operations have used generative AI for content production and audience analysis.	High
Chinese government agencies have access to a commercial ecosystem capable of supporting portions of the proposed architecture.	Moderate to high
The PLA is integrating these capabilities into military decision-support and simulation research.	High
A unified, operational LLM-driven cognitive-warfare system has been deployed under a named PLA or MSS program.	Not established
The 2023 algorithmic cognitive-warfare paper has been formally adopted as an operational directive.	Not established

The PRC research collected in this project describes a progression from psychological warfare and public-opinion warfare toward algorithmic cognitive warfare.

The central shift is from influencing an audience through information to influencing how that audience discovers, interprets, discusses, and acts upon information.

Four publications are especially important.

1.1. Intelligent communication as a cognitive-domain weapon, 2023

Chinese title: 《智能传播：认知域作战的重要场域》
Author: 李晓阳, Li Xiaoyang
Publication: PLA Daily, April 18, 2023.

Original Ministry of National Defense publication

This publication, previously translated and analyzed in the project, describes intelligent communication as an important environment for cognitive-domain operations.

Its principal concepts include algorithmic control of information exposure, social-media bots, synthetic media, audience profiling, information cocoons, emotional manipulation, and behavioral influence.

The article treats modern communication platforms as more than distribution infrastructure. Their algorithms become mechanisms capable of influencing the cognitive processes through which audiences form judgments.

Intelligence significance: The article provides a PLA-affiliated conceptual justification for exploiting commercial information systems as instruments of cognitive influence.

1.2. Algorithmic cognitive warfare, 2023

Chinese title: 《算法认知战的致效机理与关键技术探析》
Authors: Chen Changxiao, Li Hao, Chao Shuai, Feng Mingyue, and Yang Yanfei.
Institution: PLA Air Force Early Warning Academy.
Publication: Information Security and Communications Secrecy, Issue 12, 2023.

Full-text Chinese reproduction

This is the central doctrinal reference in the project.

The paper describes six stages of algorithmic cognitive warfare:

Stage	Chinese terminology	Operational purpose
1	用户画像	Profile individuals, communities, and their psychological characteristics.
2	吸引注意	Capture attention using information tailored to audience interests and psychological needs.
3	暗示参考	Shape the reference points through which audiences interpret events.
4	诱导反应	Encourage reactions and migration toward communities that reinforce selected narratives.
5	适时干预	Adjust information exposure and intervene in developing group behavior.
6	监督满足	Monitor group cohesion, satisfaction, and the persistence of desired cognitive effects.

The paper also identifies five enabling technologies: multi-source information fusion, knowledge-driven data mining, deepfakes, human-machine information filtering and recommendation, and generative large models.

Its most consequential proposition is that generative models can make information appear increasingly natural, personalized, and trustworthy while influencing audiences through repeated interaction.

Intelligence significance: This paper provides a coherent theoretical architecture for sustained, adaptive influence operations. It does not demonstrate that the architecture was implemented.

1.3. Large language models for cognitive warfare in cyberspace, 2024

Chinese title: 《面向网络空间认知战的大语言模型：技术与挑战》
Authors: Yang Guangfei and colleagues.
Publication: Journal of Command and Control, Volume 10, Issue 6, 2024, pp. 643–652.

The project’s earlier analysis identified a proposed LLM-enabled cognitive-warfare cycle connecting multi-source collection, situational understanding, decision support, information operations, and assessment.

The publication discusses the application of LLMs to cognitive confrontation and the technical challenges involved.

Its importance is the transition from algorithms supporting individual functions to a more integrated model in which language models assist with multiple phases of the information-operation cycle.

Intelligence significance: The publication connects the 2023 algorithmic-warfare concept to LLM-centered architectures. This is doctrinal and technical convergence, not evidence of a fielded system.

1.4. Cognitive warfare enters the LLM-agent era, 2026

Chinese title: 《认知战迈向大模型驱动的认知智能体阶段》
Publication: Chinese Institute of Command and Control, May 21, 2026.
Affiliations: Northwestern Polytechnical University and the CICC Cognitive and Behavioral Committee.

Original Chinese publication

This is one of the most significant subsequent publications identified in the project.

It proposes moving beyond individual AI-generated messages toward persistent cognitive agents capable of maintaining identities, interacting with audiences, coordinating activity, retaining memory, and adapting strategies.

The proposed architecture includes persistent digital identities, long-term memory, planning, multi-agent coordination, cognitive simulation, and feedback-driven adaptation.

Intelligence significance: This publication provides a conceptual bridge between algorithmic influence theory and autonomous or semi-autonomous information operations.

It remains a research proposal rather than proof of operational deployment.

Two additional Chinese papers already collected in this project are important because they move beyond the production of propaganda content into the modeling of human behavior and social systems.

2.1. PLA Academy of Military Science: Multi-agent social-network simulation

Chinese title: 《大模型驱动的社交网络多智能体仿真综述》
English: A Review of Large-Model-Driven Multi-Agent Social Network Simulation
Authors: Li Jiting, Sun Yi, Wang Yirong, Lin Yiqin, Jia Jun, and Ding Gangsong.
Institution: PLA Academy of Military Science (军事科学院).
Publication: Journal of System Simulation, Volume 38, Issue 2, February 2026.

Journal publication

This paper is especially significant because the research originates within a PLA military research institution.

The authors describe LLM-driven social-network simulation across three levels.

Simulation level	Research objective	Potential relevance to cognitive warfare
Micro	Model individual cognition, personality, decisions, and behavior.	Estimate how different audience profiles might interpret narratives.
Meso	Model interpersonal interaction, relationships, and group dynamics.	Study persuasion, social reinforcement, polarization, and information diffusion.
Macro	Model collective behavior and emergent social-network phenomena.	Examine how information campaigns might affect larger populations.

The research discusses generative agents capable of maintaining memories, conducting conversations, developing relationships, and exhibiting emergent collective behavior.

It also identifies future research priorities involving real-world social-network data, improved simulation fidelity, larger populations of agents, and longer simulation periods.

The authors acknowledge substantial limitations involving model reliability, hallucinations, social-network complexity, computational costs, and the difficulty of validating simulated behavior against actual human behavior.

Assessment: This establishes PLA institutional interest in LLM-enabled social simulation. Such simulation could support cognitive-warfare planning and assessment, but the paper itself is not evidence that the Academy of Military Science operates a deployed disinformation system.

2.2. National Defense University and Fudan University: Social cognition simulation

Chinese title: 《基于大语言模型智能体的社会认知模拟》
English: Social Cognition Simulation Based on Large Language Model Agents
Authors: Zhang Mingxin and colleagues.
Institutions: PLA National Defense University College of Politics and Fudan University.
Publication: Journal of System Simulation, February 2026.
DOI: 10.16182/j.issn1004731x.joss.25-0612.

Original academic publication

The research examines LLM agents as substitutes for or supplements to conventional rule-based agents in social simulation.

Its discussion includes social cognition, human interaction, collective behavior, decision-making, and military simulation.

The relevance to cognitive warfare is the possibility of creating simulated populations whose members respond differently to information, events, and social interactions.

Such systems could eventually support the evaluation of influence campaigns before those campaigns are conducted against real populations.

However, the paper also recognizes that LLM-generated social behavior may be unreliable, inconsistent with actual social conditions, and difficult to validate.

Assessment: The institutional affiliation and research subject establish military interest in LLM-based social cognition. They do not demonstrate an operational influence campaign.

2.3. Why the two simulation papers matter together

The Academy of Military Science publication addresses social-network simulation at the individual, group, and population levels. The National Defense University-associated publication examines the use of LLM agents to reproduce aspects of human social cognition and decision-making. The 2026 CICC cognitive-agent publication then proposes systems capable of persistent interaction and coordinated cognitive influence.

Taken together, these publications describe complementary research directions:

Simulated human behavior → Simulated social interaction → Cognitive-strategy evaluation → Persistent AI agents → Adaptive information operations.

This is an analytical synthesis of the research, not a documented development sequence or a confirmed joint program. Nevertheless, it identifies an important capability area: the potential convergence of social simulation and autonomous influence agents.

The project’s research identifies a distinction between broad Chinese political direction, military doctrine, and specific technical procurement.

3.1. Central political direction

Chinese leadership has repeatedly emphasized using AI to improve information production, distribution, reception, and feedback, while strengthening the ability to guide public opinion. The 2023 algorithmic cognitive-warfare paper explicitly invokes Xi Jinping’s direction to apply AI across the news-production and distribution process.

This provides political context for the research. It does not mean that every commercial AI system developed under this policy is intended for covert influence operations. China’s public-opinion guidance system also has legitimate administrative and communications functions, alongside its censorship, propaganda, and political-control functions.

3.2. PLA doctrine: Preparing to fight five cognitive battles

Chinese title: 《瞄准未来战争打好认知“五仗”》
Publication: PLA Daily, October 8, 2022.

Original Ministry of National Defense publication

This article, preserved in the project files, advocates preparing cognitive operations before conventional military conflict begins. It emphasizes narrative initiative, the manipulation of an adversary’s perceptions, cognitive deterrence, and coordination with other military operations.

Particularly relevant is its call to accelerate the construction of cognitive offense-and-defense theory repositories, databases, model repositories, case repositories, and operational-method repositories. It also advocates integrating neural-network systems, AI applications, and cognitive decision-support technologies.

Intelligence significance: This is evidence of a publicly articulated military requirement for technical infrastructure supporting cognitive operations.

3.3. Hybrid warfare and the civilian support ecosystem

The project also contains the Chinese military publication:

《混合战争视野下的认知域作战》
Cognitive-Domain Operations from the Perspective of Hybrid Warfare

The article argues that cognitive-domain operations are not limited to uniformed military personnel. It identifies political organizations, news media, companies, research institutions, social organizations, online personalities, and other civilian actors as potential participants. It also discusses the need for expertise spanning military affairs, communications, psychology, journalism, and computer science.

Intelligence significance: This helps explain why Chinese cognitive-warfare research and capabilities may appear across military academies, civilian universities, commercial technology companies, and government agencies rather than within one identifiable PLA organization.

The following entries consolidate the named development efforts identified during the project’s previous investigations.

Evidence classifications: A procurement notice establishes a stated acquisition requirement. A corporate announcement establishes a claimed development effort. Neither independently proves operational deployment.

Program or system	Organization	Identified capability	Relevance	Status
混合智能认知分析与决策平台 (Hybrid Intelligent Cognitive Analysis and Decision Platform)	Zhongke Fanyu (Wuhan), associated developer	Human-machine cognitive analysis and decision support	Explicit military cognitive-domain requirements in a project-registration listing	Developer association identified; military contract unverified
基于多智能体的认知引导系统 (Multi-Agent Cognitive Guidance System)	Beijing Jiaotong University	Multi-agent cognitive-guidance system	Potential influence or behavioral-guidance technology	Procurement notice identified; mission and contractor unresolved
智能中枢典型应用项目 (Intelligent Hub Typical Applications Project)	PLA procurement system	LLM training management, knowledge services, and decision support	Potential supporting infrastructure	Procurement requirement identified
某信息域智能博弈兵推模型生成演化环境 (Information-Domain Intelligent Wargaming Model Generation and Evolution Environment)	PLA procurement system	Red/blue agents, reinforcement learning, model evolution	Relevant to agent experimentation and simulation	Procurement requirement identified
基于智能体的信息推荐软件 (Agent-Based Information Recommendation Software)	PLA procurement listing	Agent-based information recommendation	Potential relevance to tailored information delivery	Demonstration application; cognitive-warfare mission unverified
基于大模型的图像生成软件 (LLM-Based Image Generation Software)	PLA procurement listing	Synthetic-image generation	Potential relevance to AI-generated influence content	Demonstration application; deceptive use unverified
先进认知战 (Advanced Cognitive Warfare)	Xianjin AI	Commercial cognitive-warfare product	Explicitly marketed for cognitive-warfare applications	Product identified; military customer unverified
GoPro	GoLaxy / Zhongke Tianji	AI-assisted social-media analysis and influence capabilities described in external research	Strong relevance to audience profiling and influence	Product and associated reporting identified; specific deployment requires corroboration
Army Dog Smart Online Commenter Management System	Zhongke Dianji	Coordination of online-commenting tasks	Direct relevance to organized narrative amplification	Documented platform; LLM autonomy not established

4.1. Hybrid Intelligent Cognitive Analysis and Decision Platform

Chinese: 混合智能认知分析与决策平台
Associated company: 中科凡语（武汉）, Zhongke Fanyu (Wuhan).
Associated executive: 周伟枭, Zhou Weixiao.

The project identified a July 2025 registration listing that explicitly referenced military cognitive-domain operational requirements. Separate provincial reporting associated Zhongke Fanyu with a platform of the same name and an AI decision-making product.

Hubei Daily reporting

This is one of the strongest potential research-to-application connections in the project because the requirement uses explicit military cognitive-domain terminology. However, the evidence does not yet establish a named PLA contracting authority, contract number, classified or unclassified program sponsor, or operational deployment.

4.2. Beijing Jiaotong University multi-agent cognitive-guidance procurement

Chinese: 北京交通大学基于多智能体的认知引导系统建设项目

The October 27, 2026 procurement announcement identified a budget of approximately RMB 230,000 and a November 9 bid deadline.

Procurement listing

The technical specification is particularly important because the phrase cognitive guidance could refer to several applications, including education, behavioral research, or influence technology. Without the specification, it would be inappropriate to classify the project as a military disinformation platform.

4.3. PLA intelligent war-gaming environment

The previously identified procurement for an information-domain intelligent war-gaming environment specified red/blue agents, reinforcement learning, and agent evolution.

The reported budget was RMB 3.4 million.

PLA procurement notice

This establishes interest in multi-agent experimentation within military information-domain systems.

4.4. PLA Intelligent Hub project

The January 2025 procurement describes LLM-related management, knowledge services, and AI decision-support functions.

PLA Intelligent Hub procurement

The significance lies in its potential to provide common AI infrastructure for military applications.

The commercial ecosystem is central to understanding how the PRC could translate LLM research into operational information-warfare capabilities. The most relevant firms fall into three categories: audience intelligence and profiling, information generation and distribution, and military-oriented AI decision support.

5.1. GoLaxy / Zhongke Tianji

Chinese: 中科天玑
Identified platform: GoPro
Other identified products: 网络社会认知大脑 (Network Society Cognitive Brain) and 全要素AI舆情 (All-Factor AI Public Opinion).

GoLaxy product information

GoLaxy is among the highest-priority companies in this investigation.

Its public-facing technologies concern large-scale information collection, public-opinion analysis, and the modeling of social-media environments. Separately, research based on leaked company materials has described GoPro capabilities relevant to profiling social-media users and supporting AI-assisted influence activity.

The overlap with the 2023 PLA framework is substantial.

PLA cognitive-warfare function	Relevant GoLaxy capability
User profiling	Social-media collection, audience analysis, and user characterization
Attention capture	Identification of topics and audience interests
Reference shaping	Analysis of narratives and information environments
Induced reactions	Potential support for targeted engagement
Timely intervention	Monitoring changing public-opinion conditions
Feedback	Assessment of online discussion and audience responses

The first three relationships are particularly consistent with the company’s documented public-opinion technology. The latter three require more detailed evidence about the specific GoPro implementation and its actual users.

Assessment: GoLaxy represents one of the closest identified commercial parallels to the PLA’s algorithmic cognitive-warfare model. However, the project’s evidence does not establish that GoPro was commissioned to implement Chen Changxiao’s six-stage framework.

5.2. Xianjin AI

Chinese: 先进智能
Product: 先进认知战 (Advanced Cognitive Warfare).

Official company website

This company is noteworthy because its product terminology explicitly references cognitive warfare. That makes it a stronger mission-oriented lead than a general-purpose LLM provider or public-opinion monitoring company.

The remaining intelligence questions concern its customers, underlying model architecture, supported data sources, deployment environment, and whether the product provides operational influence capabilities or primarily analysis and simulation.

Assessment: Explicit commercial cognitive-warfare positioning. A direct PLA or MSS operational relationship remains unverified.

5.3. Utenet / Yuanting Technology

Chinese: 渊亭科技
Identified systems: Tianji military large model, Tianyan simulation platform, and Tianchou decision-optimization system.

Tianji military large-model announcement
Tianyan simulation platform
Tianchou decision optimization

Utenet’s military AI products are relevant to the broader transition toward LLM-assisted military planning, knowledge processing, and simulation. The project’s earlier research also identified its 2024 Military Large Model Evaluation System White Paper. The significance is primarily architectural: military-oriented LLMs, simulation, and decision-support capabilities could provide components useful in a larger cognitive-warfare system.

Assessment: Relevant military AI development. No verified evidence that these particular products are conducting disinformation or covert influence operations.

5.4. Other companies identified

The project also identified several companies associated with public-opinion monitoring, online information management, or influence-related technologies.

Company	Identified relevance	Principal limitation
Warming High-Tech (沃民高新)	Public-opinion analysis and reported government-linked influence ecosystem	Specific LLM-driven covert operation unverified
Zhongke Dianji (中科点击)	Online-commentator coordination and information monitoring	Autonomous LLM capability unverified
Meiya Pico (美亚柏科)	Data analysis and digital-investigation capabilities; named in Taiwanese intelligence allegations	Specific cognitive-warfare contract unverified
TRS Information Technology	Information-processing and online-commentator technology	Government customer relationships do not prove product-specific use
Eefung	Public-opinion technology and reported defense-university research cooperation	Operational cognitive-warfare role unverified
Baizhang Technology (百章科技)	Previously identified adjacent AI/information-analysis lead	Insufficient project evidence to establish a specific influence capability
Qi-Anxin (奇安信)	Cybersecurity and threat-intelligence technology	No demonstrated link between the reviewed products and an LLM disinformation operation
iFlytek (科大讯飞)	Chinese-language AI and speech-generation capabilities	General-purpose AI capability is not evidence of cognitive-warfare deployment

The April 2026 German Marshall Fund investigation Outsourcing Influence is particularly useful for examining Chinese online-commentary contractors and the relationship between commercial technology and state-directed information management.

Outsourcing Influence, April 2026 (PDF)

The broader finding is that China’s influence-technology ecosystem is not confined to military contractors. It includes firms serving civilian government agencies, public-security organizations, propaganda functions, and commercial customers.

The evidence collected across the project identifies several Chinese state organizations with actual, alleged, or potential roles in AI-enabled cognitive operations.

Their relationships must be distinguished carefully.

Organization	Chinese designation	Relevant role	Evidence status
PLA Political Work Department	中央军委政治工作部	Military political work and potential psychological/information operations	Institutional role established; specific LLM platform unverified
PLA Academy of Military Science	军事科学院	Military research into LLM-based social simulation	Direct academic evidence
PLA National Defense University	国防大学	Military education and research into social cognition and cognitive operations	Direct academic evidence
PLA Air Force Early Warning Academy	空军预警学院	Algorithmic cognitive-warfare research	Direct academic evidence
Ministry of State Security	国家安全部	Foreign intelligence and security activities; alleged involvement in influence-related commercial ecosystem	Specific LLM influence platform unverified
Ministry of Public Security	公安部	Public-security information operations and documented links to Chinese influence activity	Organizational links reported; program-level evidence varies
Cyberspace Administration of China	国家互联网信息办公室	Online information governance, platform regulation, and public-opinion management	Institutional role established
Central Publicity Department	中共中央宣传部	Party propaganda and information guidance	Institutional role established
Chinese Institute of Command and Control	中国指挥与控制学会	Publication and dissemination of military AI and cognitive-agent research	Research dissemination established; not proof of operational command
Former PLA Base 311	311基地	Historically associated with political and psychological warfare directed particularly toward Taiwan	Historical role documented; current command relationships require verification

The Ministry of State Security question

The project has not established a direct documentary chain showing the MSS commissioning a named LLM cognitive-warfare platform for foreign disinformation operations. Taiwanese intelligence assessments have alleged involvement by Chinese intelligence, public-security, and political-work organizations in an ecosystem using commercial technology for information collection and influence. Those allegations are relevant collection leads, but they should not be converted into verified contract relationships without corroborating Chinese records.

The PLA organizational question

The PLA’s 2024 restructuring complicates the attribution of older cognitive-warfare organizations to current military command structures. Historical references to the Strategic Support Force, its network and psychological-warfare elements, and Base 311 cannot automatically be mapped onto the current Information Support Force or other successor organizations.

Assessment: The evidence supports a distributed military, intelligence, government, and commercial research ecosystem. It does not support describing all identified organizations as participants in one centrally managed LLM cognitive-warfare program.

7.1. Spamouflage / Dragonbridge

Category: Coordinated inauthentic influence activity.
Attribution: Chinese-linked; specific organizational attribution varies by reporting source.

Spamouflage is among the most extensively documented Chinese-linked online influence operations. Its activity has involved networks of inauthentic social-media accounts, coordinated posting, political narratives, and attempts to influence audiences outside mainland China. A particularly important finding in the project’s uploaded research concerns the use of ChatGPT by operators associated with Spamouflage.

OpenAI’s 2024 reporting described the use of ChatGPT to summarize and analyze the sentiment of posts by overseas Chinese dissidents who were targets of harassment. This is important because it demonstrates a use of generative AI for audience analysis, not merely for generating propaganda. That distinction aligns directly with the first stage of the PLA’s algorithmic cognitive-warfare model: user profiling and cognitive assessment.

7.2. AI-generated content in Chinese-linked influence campaigns

Public threat-intelligence reporting has described Chinese-linked actors using generative AI to assist with political messaging, multilingual content, images, and synthetic media.

Such activity can support several distinct operational functions.

Observed or reported function	Relevance to LLM cognitive warfare
Drafting political commentary	Reduces the cost of producing influence content
Rewriting messages for different audiences	Supports narrative localization
Translating and adapting material	Expands multilingual reach
Producing synthetic images or video	Expands deceptive or emotionally persuasive content options
Summarizing social-media conversations	Supports audience analysis
Analyzing sentiment	Supports monitoring and feedback
Coordinating inauthentic accounts	Supports amplification, although LLM autonomy is not necessarily involved

The important limitation is that using generative AI does not, by itself, establish autonomous cognitive warfare. An operator using an LLM to draft posts remains a human-directed influence operation. A system independently maintaining personas, selecting targets, interacting with audiences, evaluating reactions, and revising its strategy would represent a substantially more advanced capability. The project has evidence of the former and research proposing the latter.

7.3. Taiwan as a principal operational environment

Taiwan is one of the most important environments for examining whether China’s cognitive-warfare doctrine is reflected in actual operations. The project’s previous research identified Taiwanese intelligence reporting on Chinese-linked disinformation activity during 2025 and 2026. The January 2026 Taiwanese National Security Bureau assessment reported approximately 45,590 inauthentic account sets and 2.31 million pieces of disinformation during 2025.

Later reporting described increasing volumes of disputed or false content during 2026, including narratives concerning defense, energy, food security, and elections. These figures are Taiwanese intelligence assessments and should be attributed accordingly.

Operational alignment

The reported Taiwan campaigns exhibit several characteristics consistent with the Chinese cognitive-warfare literature:

• Localized narratives: Content addresses specific political, economic, and social concerns rather than relying exclusively on broad ideological messaging.

• Audience segmentation: Different narratives can be directed toward communities with different interests and vulnerabilities.

• Social amplification: Inauthentic accounts and coordinated networks can increase the apparent popularity or credibility of selected narratives.

Institutional distrust: Campaigns can exploit disputes involving government competence, national defense, public safety, and social stability.

Persistent activity: Influence operations may continue outside election periods and conventional military crises.

7.4. Why actual operations matter to the doctrinal assessment

The strongest operational evidence is not the existence of AI-generated posts. It is the convergence of three observed or documented activities. These activities correspond to significant portions of the PLA’s proposed cognitive-warfare architecture. What remains unverified is whether Chinese operators have integrated them into an autonomous, adaptive system that conducts the entire influence cycle without continuous human direction.

The accumulated research supports the following analytical model of how Chinese LLM-enabled cognitive warfare could operate.

This is a synthesis of the project’s sources, not a recovered PLA system diagram.

Layer	Function	Relevant technologies	Supporting evidence
1. Collection	Gather information about audiences and their information environments	OSINT, social-media monitoring, data fusion	2023 PLA paper, GoLaxy research
2. Cognitive modeling	Characterize beliefs, interests, behavior, and social relationships	User profiling, clustering, association mining, LLM analysis	2023 paper, 2024 LLM paper
3. Social simulation	Examine potential individual and collective responses	Generative agents, social-network simulation, multi-agent modeling	2026 Academy of Military Science and National Defense University research
4. Content generation	Produce contextually relevant narratives and synthetic media	LLMs, multimodal generation, translation	PLA doctrine, documented AI-assisted influence activity
5. Distribution	Place content into relevant information environments	Social platforms, recommendation systems, coordinated accounts	2023 paper, commercial influence platforms
6. Interaction	Maintain conversations and respond to audience behavior	Persistent LLM agents, memory, agent coordination	2026 CICC cognitive-agent research
7. Assessment	Monitor changes in discussion, engagement, and sentiment	Public-opinion analytics, sentiment analysis, feedback systems	2023 paper, Spamouflage AI-use reporting
8. Adaptation	Revise subsequent activity according to observed responses	Decision support, planning agents, simulation	Proposed in 2024–2026 research; operational integration unverified

The central technological change

Traditional coordinated inauthentic behavior generally depends on people or scripts managing accounts, distributing messages, and measuring engagement. The Chinese research describes a more ambitious architecture in which LLMs could support interpretation, planning, content generation, interaction, and evaluation within the same operational environment.

Multi-agent systems could allow different synthetic identities to exhibit distinct characteristics while pursuing a common objective. Social simulation could provide a way to evaluate hypothetical audience responses before committing resources to an information campaign. Neither capability guarantees effective persuasion. Simulated behavior may diverge substantially from real human behavior, and engagement metrics are not reliable evidence of changed beliefs.

The strategic objective described in the doctrine is not simply to increase message volume. It is to influence the process through which populations construct and revise their understanding of events.

Three categories need to remain separate in this investigation.

• Misinformation is inaccurate or misleading information circulated without necessarily intending to deceive.

• Disinformation is deliberately false or misleading information disseminated to deceive.

• Cognitive warfare is broader. It seeks to influence perceptions, judgments, emotions, and decisions. It may use false information, selective presentation of true information, fabricated identities, emotional narratives, and manipulation of the information environment.

Chinese military publications frequently discuss influence mechanisms without limiting them to demonstrably false claims. This distinction matters because an LLM-enabled cognitive operation might rely primarily on authentic information selectively framed and distributed to exploit existing disagreements. Consequently, a collection strategy focused exclusively on detecting false AI-generated content would miss important parts of the proposed capability. The relevant indicators include coordinated artificial identities, concealed sponsorship, systematic audience profiling, unusual patterns of personalized engagement, and evidence that information exposure is being deliberately manipulated.

Period	Development	Significance
2022	PLA publications discuss algorithmic cognitive operations and preparing cognitive campaigns before conflict.	Establishes military doctrinal interest.
April 2023	Li Xiaoyang publishes Intelligent Communication: An Important Arena of Cognitive-Domain Operations.	Connects intelligent communication with cognitive warfare.
2023	PLA Air Force Early Warning Academy publishes the six-stage algorithmic cognitive-warfare framework.	Provides a structured conceptual model.
2023	PLA procurement describes an intelligent information-domain wargaming environment.	Establishes investment in agent-based military experimentation.
2024	Research on LLMs for cyberspace cognitive warfare appears in the Journal of Command and Control.	Extends the concept toward LLM-enabled processes.
2024	The original PLA research team publishes additional work on cognitive-domain operational forms.	Demonstrates research continuity.
2024	Public threat reporting documents generative-AI assistance in Chinese-linked influence activity.	Establishes observed AI use in influence operations.
January 2025	PLA Intelligent Hub procurement identifies LLM and decision-support requirements.	Establishes relevant military AI acquisition activity.
July 2025	Hybrid Intelligent Cognitive Analysis and Decision Platform appears in project-registration reporting.	Explicit military cognitive-domain requirement, with an associated commercial developer.
February 2026	Academy of Military Science and National Defense University-associated researchers publish LLM social-simulation research.	Demonstrates military research into agent-based modeling of human and collective behavior.
May 2026	CICC publishes research on LLM-driven cognitive agents.	Describes persistent, adaptive, multi-agent cognitive operations.
2026	Taiwanese intelligence reports extensive Chinese-linked influence activity and increasing AI involvement.	Provides operational context, subject to attribution and measurement limitations.
October 2026	Beijing Jiaotong University advertises a multi-agent cognitive-guidance system procurement.	Creates a new technical-specification and contractor-identification lead.

Assessment of the timeline: The publications show increasingly sophisticated technical concepts. The procurement records show investment in some related technologies. The operational reporting establishes that Chinese-linked influence activity is ongoing.

The project’s most consequential unresolved questions concern integration, attribution, and operational effectiveness.

Priority	Intelligence requirement	Evidence needed
Critical	Has the PLA formally adopted an LLM-enabled cognitive-warfare architecture?	Operational requirements, approved implementation plans, or authoritative military directives
Critical	Is a Chinese intelligence organization operating persistent LLM personas against foreign audiences?	Authenticated operational records, platform telemetry, or independently corroborated investigative reporting
Critical	Which contractors are supplying operational cognitive-warfare systems?	Contracts, award notices, acceptance documents, and customer references
High	Does GoLaxy’s GoPro platform support coordinated autonomous influence activity?	Technical documentation, verified deployment records, and corroborated operator evidence
High	What is the intended mission of Beijing Jiaotong University’s cognitive-guidance system?	Complete procurement specification and subsequent award
High	Does Zhongke Fanyu’s platform have a confirmed PLA customer?	Original project-registration documentation and contractual evidence
High	Are LLM social simulations being used to evaluate actual influence campaigns?	Military evaluation reports, test records, and simulation-to-operation documentation
High	Can observed Chinese-linked campaigns be traced to a specific military or intelligence sponsor?	Command relationships, authenticated tasking, and contractor attribution
High	How effective are the reported influence operations?	Independent behavioral measurements rather than post volume or engagement counts

Most useful Chinese search terminology

Future collection should retain the original terminology used in Chinese publications and procurement records.

Chinese term	English meaning
算法认知战	Algorithmic cognitive warfare
认知域作战	Cognitive-domain operations
认知智能体	Cognitive agent
大模型驱动	Large-model-driven
多智能体协同	Multi-agent coordination
社会认知模拟	Social cognition simulation
社交网络多智能体仿真	Multi-agent social-network simulation
认知引导系统	Cognitive-guidance system
智能舆论引导	Intelligent public-opinion guidance
认知攻防	Cognitive offense and defense
智能传播	Intelligent communication
计算宣传	Computational propaganda
舆情态势感知	Public-opinion situational awareness
认知对抗推演	Cognitive-confrontation simulation
认知域作战需求	Cognitive-domain operational requirements
项目验收	Project acceptance
中标公告	Contract-award announcement

The following are the most important documents underpinning the consolidated assessment.

Document	Institution or publisher	Intelligence value
《算法认知战的致效机理与关键技术探析》, 2023	PLA Air Force Early Warning Academy	Foundational six-stage framework
《智能传播：认知域作战的重要场域》, 2023	PLA Daily	Intelligent communication and cognitive operations
《面向网络空间认知战的大语言模型：技术与挑战》, 2024	Journal of Command and Control	LLM-enabled cognitive-warfare architecture
《认知域作战主要样式演进与发展趋势》, 2024	National Defense Technology	Continuity of PLA cognitive-warfare research
《大模型驱动的社交网络多智能体仿真综述》, 2026	PLA Academy of Military Science	Multi-agent social simulation
《基于大语言模型智能体的社会认知模拟》, 2026	National Defense University-affiliated researchers	LLM social cognition simulation
《认知战迈向大模型驱动的认知智能体阶段》, 2026	Chinese Institute of Command and Control	Persistent cognitive-agent architecture
Decoding China’s AI-Powered Algorithmic Cognitive Warfare	Independent research report in project files	Synthesis of Chinese doctrine and implications
Outsourcing Influence, 2026	German Marshall Fund	Chinese commercial influence infrastructure

The project also contains Chinese military publications on hybrid warfare and cognitive-domain operations, including Preparing to Fight Five Cognitive Battles in Future Warfare.

Source limitations: This synthesis draws on the material available in the PRC project and prior investigations. It is not a fresh, exhaustive verification of every public claim or procurement status as of November 3, 2026. Some procurement notices and corporate associations remain leads rather than confirmed operational relationships.

Chinese military researchers are developing a form of cognitive warfare that uses large language models (LLMs) to influence how people think, interpret events, and make decisions. Rather than simply generating propaganda, these systems would combine intelligence collection, psychological profiling, synthetic content, and social-media engagement to shape public opinion and manipulate behavior.

A 2023 paper from the PLA Air Force Early Warning Academy outlines a six-stage framework for this approach. Subsequent research from the Academy of Military Science, National Defense University, and Chinese Institute of Command and Control explores how LLMs and autonomous agents could simulate human behavior, maintain online identities, and adapt messaging based on audience reactions.

Chinese technology companies are developing many of the capabilities described in this research, including social-media monitoring, behavioral analysis, synthetic-media generation, and automated information distribution. Military procurement records also show investment in LLM infrastructure, multi-agent systems, and AI-assisted decision-making. Meanwhile, Chinese-linked influence operations have already used generative AI to produce content and analyze target audiences.

Taken together, these developments suggest China is moving beyond traditional propaganda toward more automated and personalized cognitive warfare. What remains unclear is how much of this technology has been integrated into actual military or intelligence operations. There is still no verified evidence of a deployed system combining all six stages of the PLA framework under a named military or intelligence command. The research and investment demonstrate a clear direction, but the existence and effectiveness of a fully operational system remain unconfirmed.

Most significant intelligence judgment

The most significant development in Chinese cognitive warfare is not the ability to produce more convincing deepfakes or larger amounts of propaganda. It is the potential to combine mass data collection, simulated human behavior, persistent online identities, personalized messaging, and continuous feedback into a system capable of conducting sustained influence operations.

Chinese military research and commercial technology development show progress across these areas. However, the evidence primarily concerns individual capabilities rather than a fully integrated system. While Chinese-linked influence operations have already demonstrated the use of generative AI, there is no verified evidence that the PLA or Chinese intelligence services have deployed an autonomous platform combining these technologies into a single operational capability.

The next important step is identifying whether these systems have moved beyond research and testing. A Chinese military or government contract, technical specification, or acceptance report identifying a cognitive-warfare platform, its sponsoring organization, and its developer would provide stronger evidence of formal adoption. Until then, China’s LLM-enabled cognitive warfare should be assessed as an emerging capability supported by military research, commercial development, and demonstrated AI-assisted influence activity, rather than a confirmed fully operational system.