The People’s Liberation Army (PLA) is actively exploring the use of generative AI, particularly large language models (LLMs), to enhance military intelligence tasks such as data processing, analysis, and decision-making. While recognizing potential benefits, the PLA is also aware of the risks and challenges, including ideological biases and the threat of misinformation from foreign counterintelligence operations. #PLA #DeepSeek #OpenAI #Meta
Keypoints
- The PLA and China’s defense industry are developing and adapting generative AI models, including proprietary and open-source LLMs from both domestic and foreign sources, to support intelligence activities.
- Patent filings show the PLA has designed generative AI systems to assist in intelligence processes such as OSINT generation, satellite image analysis, and event extraction.
- Academic research affiliated with PLA institutions emphasizes both the transformative potential of generative AI for intelligence and its limitations such as hallucinations and ideological bias.
- Generative AI is expected to enhance national defense science and technology intelligence by improving data comprehension, scenario generation, and report synthesis, but it also introduces risks like fake documents and deepfakes.
- The PLA is closely monitoring the US military’s application of generative AI to gain insights and adopt similar best practices.
- Concerns persist about the PLA’s use of AI models influenced by CCP ideology potentially reducing objectivity, and about foreign adversaries leveraging generative AI to spread disinformation.
- Non-PLA entities within China’s party-state system also warn about ideological challenges posed by Western-developed AI tools like ChatGPT affecting intelligence work.
MITRE Techniques
- [T1593] Open-Source Intelligence Gathering – The PLA uses generative AI to generate OSINT products and process large amounts of open-source data (“…use OSINT, human intelligence (HUMINT), signals intelligence (SIGINT), geospatial intelligence (GEOINT), and technical intelligence (TECHINT) data to train a military LLM…”).
- [T1074] Data Staged – AI models process satellite imagery and event data to facilitate intelligence tasks (“…processing satellite imagery, supporting event extraction, and processing event data…”).
- [T1609] Reconnaissance – Generative AI assists in evaluating and analyzing battlefield intelligence to aid decision-making and planning (“…integrate large amounts of battlefield intelligence into a comprehensive battlefield situation report automatically…”).
- [T1486] Data Encrypted for Impact (Implied) – Awareness of deepfake technology emphasizes the risk of generating convincing false information affecting intelligence integrity (“…warns deepfakes could be used to interfere with intelligence work…”).
Indicators of Compromise
- [File Hashes] Generative AI models and tools – Examples include proprietary models like DeepSeek’s LLM used by PLA, and foreign LLMs such as Meta’s Llama 13B and OpenAI’s ChatGPT.
- [Domains] AI and defense contractor platforms – Domains associated with Chinese defense contractors providing generative AI-based OSINT models to PLA.
- [Patent Applications] PLA and Chinese defense industry patents – Filed methods and systems describing the use of generative AI in intelligence workflows, including satellite imagery processing and event data analysis.
Read more: https://www.recordedfuture.com/research/artificial-eyes-generative-ai-chinas-military-intelligence