Ajeet Kumar

AttackLogGen: Benchmarking LLMs for generating attack logs

Evaluation across semantic similarity, distributional consistency, discriminability tests, and structural fidelity shows that LLM-generated logs—particularly using GPT-4o-mini, closely approximate real MHMS activity, enabling safe and scalable experimentation for intrusion-detection research.