Full Report
Explore how China’s PLA is adopting generative AI for military intelligence. This Insikt Group report reveals AI-driven intelligence tools, strategic adaptations, and implications for global security.
Analysis Summary
# Threat Actor: People’s Liberation Army (PLA) and China's Defense Industry
## Attribution & Identity
The actor is identified as the **People’s Liberation Army (PLA)** and associated entities within **China’s defense industry**. The context indicates a state-sponsored effort to integrate emergent technologies into intelligence operations.
## Activity Summary
The PLA and China's defense industry are actively demonstrating clear interest in, designing methods for, and likely procuring generative artificial intelligence (AI) to support intelligence work. This includes adapting foreign and domestic Large Language Models (LLMs) to create specialized models for intelligence tasks such as processing and analyzing data, generating intelligence products, facilitating early warning, and supporting decision-making. They are adapting generalized models for intelligence-specific applications.
## Tactics, Techniques & Procedures
- Generating open-source intelligence (OSINT) products using generative AI.
- Processing satellite imagery using generative AI systems.
- Supporting event extraction and processing of event data via AI systems.
- Training specialized military LLMs using a fusion of intelligence data types: OSINT, HUMINT, SIGINT, GEOINT, and TECHINT.
- **Potential Counterintelligence TTP:** Chinese counterintelligence organizations could apply generative AI to produce convincing inauthentic content to mislead Western intelligence analysts and degrade open-source intelligence value.
## Targeting
- **Sectors:** Military Intelligence, Intelligence Collection and Analysis (across the intelligence cycle).
- **Geography:** Global scope implied by the use of foreign LLMs and intelligence collection, but the actor is state-affiliated (China).
- **Victims:** Not explicitly named, but the primary targeting is related to improving intelligence advantage against perceived adversaries (The West).
## Tools & Infrastructure
- **Malware families used:** Not explicitly malware, but rather adapted LLMs/Generative AI models.
- **Foreign LLMs:** Models from Meta, OpenAI, and BigScience are likely adapted.
- **Domestic LLMs:** Models from DeepSeek, Tsinghua University, Zhipu AI, and Alibaba Cloud are likely adapted.
- **Specific Model Adoption:** The PLA very likely adopted **DeepSeek’s generative AI models** rapidly in early 2025, potentially via a DeepSeek-based OSINT model provided by a Chinese defense contractor.
## Implications
The PLA's integration of generative AI promises increased speed, efficiency, and scale in intelligence tasks. However, reliance on models trained under Communist Party of China (CCP) ideology risks reducing the objectivity of analysis. The most significant external implication is the risk of **degradation of Western intelligence sources** due to China using generative AI to produce highly convincing disinformation targeting open-source intelligence channels.
## Mitigations
- Intelligence communities must be vigilant against the potential for highly convincing, AI-generated inauthentic content designed to mislead Western analysts using open-source information.
- Organizations integrating AI into intelligence workflows must rigorously assess the outcomes of experiments to prevent inaccurate intelligence from degrading decision-making quality.
- Be aware of intelligence collection derived from LLMs potentially affected by underlying ideological biases if trained on state-aligned data.