A systematic review of 73 research papers reveals that large language models (LLMs) like GPT, BERT, and T5 present significant dual-use challenges, enabling innovation while simultaneously creating vulnerabilities for phishing, malicious code generation, privacy breaches, and misinformation spread. The study, published in Frontiers of Engineering Management (2025), warns that without systematic regulation and improved defense mechanisms, LLM misuse threatens data security, public trust, and social stability.
The research team from Shanghai Jiao Tong University and East China Normal University screened over 10,000 documents to identify key threats, categorizing them into misuse-based risks and malicious attacks targeting models. Misuse includes phishing emails crafted with near-native fluency, automated malware scripting, identity spoofing, and large-scale false information production. Malicious attacks occur at both data/model levels—such as model inversion, poisoning, and extraction—and user interaction levels including prompt injection and jailbreak techniques that can access private training data or bypass safety filters.
On defense strategies, the study summarizes three technical approaches: parameter processing to reduce attack exposure, input preprocessing to detect adversarial triggers, and adversarial training using red-teaming frameworks for robustness improvement. Detection technologies like semantic watermarking and CheckGPT can identify model-generated text with up to 98–99% accuracy, as detailed in the research available at https://doi.org/10.1007/s42524-025-4082-6. Despite these advances, defenses often lag behind evolving attacks, indicating an urgent need for scalable, low-cost, multilingual-adaptive solutions.
The authors emphasize that technical safeguards must coexist with ethical governance, arguing that hallucination, bias, privacy leakage, and misinformation represent social-level risks requiring more than engineering solutions. Future models should integrate transparency, verifiable content traceability, and cross-disciplinary oversight through ethical review frameworks, dataset audit mechanisms, and public awareness education to prevent misuse and protect vulnerable groups.
The secure and ethical development of LLMs will shape how societies adopt AI, with robust defense systems potentially protecting financial systems from phishing, reducing medical misinformation, and maintaining scientific integrity. Watermark-based traceability and red-teaming may become industry standards for model deployment, while future work should focus on AI responsible governance, unified regulation frameworks, safer training datasets, and model transparency reporting. If well-managed, LLMs can evolve into reliable tools supporting education, digital healthcare, and innovation ecosystems while minimizing risks linked to cybercrime and social misinformation.
