Scholarlink

<p>The rapid digitalization of agriculture driven by smart farming (SF) and precision agriculture (PA) is transforming how food is produced, managed, and distributed. At the core of this transformation lies the integration of Large Language Models (LLMs), which offer enhanced capabilities in data interpretation, decision support, and knowledge dissemination. However, this integration also introduces a new attack surface with substantial cybersecurity implications. This paper provides the first systematic security analysis of LLMs in agricultural systems, presenting a threat taxonomy grounded in the MITRE ATT&amp;CK (Adversarial Tactics, Techniques, and Common Knowledge) framework, a cybersecurity knowledge base that describes how attackers carry out intrusions. This research identifies key adversarial tactics including reconnaissance, resource development, execution, and exfiltration (unauthorized data extraction) that can exploit LLM-driven agricultural workflows. To address these risks, this study examines the effectiveness of existing mitigation strategies, such as prompt filtering, adversarial training, and encryption-based safeguards. The analysis reveals critical gaps in current defenses and highlights directions for future research, including adaptive threat detection, secure model deployment, and domain-specific alignment techniques. This work offers a foundational framework for securing LLMs in SF and PA, promoting resilient, trustworthy, and sustainable agricultural innovation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Emerging roles of large language models (LLMs) in smart farming and precision agriculture: security challenges, threat taxonomy, and mitigation strategies

  • 期刊论文

  • 发表日期：

    2026年5月26日

查看全文
  • Sarina Dastgerdy,
  • Abbas Yazdinejad,
  • Ali Dehghantanh

摘要

The rapid digitalization of agriculture driven by smart farming (SF) and precision agriculture (PA) is transforming how food is produced, managed, and distributed. At the core of this transformation lies the integration of Large Language Models (LLMs), which offer enhanced capabilities in data interpretation, decision support, and knowledge dissemination. However, this integration also introduces a new attack surface with substantial cybersecurity implications. This paper provides the first systematic security analysis of LLMs in agricultural systems, presenting a threat taxonomy grounded in the MITRE ATT&CK (Adversarial Tactics, Techniques, and Common Knowledge) framework, a cybersecurity knowledge base that describes how attackers carry out intrusions. This research identifies key adversarial tactics including reconnaissance, resource development, execution, and exfiltration (unauthorized data extraction) that can exploit LLM-driven agricultural workflows. To address these risks, this study examines the effectiveness of existing mitigation strategies, such as prompt filtering, adversarial training, and encryption-based safeguards. The analysis reveals critical gaps in current defenses and highlights directions for future research, including adaptive threat detection, secure model deployment, and domain-specific alignment techniques. This work offers a foundational framework for securing LLMs in SF and PA, promoting resilient, trustworthy, and sustainable agricultural innovation.