What AI says about AdaptiveGuard

AdaptiveGuard is a replication package for continual LLM safety alignment research, designed to study and implement adaptive guardrails to keep language models safe over time. It provides code, data, models, and scripts to reproduce experiments and evaluate defenses against jailbreak-style attacks, including data preparation steps like combining attack datasets. The project offers an environment setup guide (Python 3.12, PyTorch with CUDA) and a repository structure (src, data, models, scripts, README) to support replicating and validating the experiments.