This comprehensive survey examines the growing security concerns as LLMs become increasingly embedded in critical industries like healthcare, finance, and education.

• Attack vectors: Explores how malicious actors can implant hidden backdoors into LLMs during training or fine-tuning
• Defense mechanisms: Reviews current strategies to detect and neutralize backdoor threats in language models
• Evaluation frameworks: Analyzes methods to assess LLM vulnerability and security robustness
• Cross-industry implications: Highlights specific risks for regulated sectors where LLM deployment is expanding

As LLMs continue rapid adoption across sensitive domains, understanding these security vulnerabilities becomes essential for responsible AI deployment and governance strategies.

A Survey on Backdoor Threats in Large Language Models (LLMs): Attacks, Defenses, and Evaluations