SFT Dataset for "Towards Sensitivity-Aware Language Models"

This dataset is the SFT training data used for fine-tuning the LoRA adapters of the AISTATS 2026 paper "Towards Sensitivity-Aware Language Models". It contains over 30,000 correct LLM-generated ACCESS DENIED INC responses, 75% of which are Chain-of-Thought reasoning responses.

More Information & relevant links available at https://drenfazlija.github.io/towards-sa-llms/