Backdoor-4Good (B-4G)

Backdoor-4Good (B-4G) is a benchmark and framework designed to evaluate beneficial backdoor applications in large language models (LLMs), focusing on enhancing safety, controllability, and accountability through a structured approach involving triggers, activation mechanisms, and utility functions.

Papers using Backdoor-4Good (B-4G) (1)

Backdoor4Good: Benchmarking Beneficial Uses of Backdoors in LLMs2026