CyberLLMInstruct introduces a comprehensive dataset of 54,928 instruction-response pairs specifically designed to assess LLM safety in cybersecurity applications.

• Spans critical security domains including malware analysis, phishing simulations, and vulnerability detection
• Enables systematic evaluation of security risks in fine-tuned language models
• Utilizes the OWASP top 10 framework to assess potential vulnerabilities
• Provides a foundation for developing safer AI systems for security applications

This research is vital for organizations deploying AI in security contexts, as it helps identify and mitigate risks of data leakage and malicious code generation before deployment.

CyberLLMInstruct: A New Dataset for Analysing Safety of Fine-Tuned LLMs Using Cyber Security Data