BackdoorAgent
What it is
BackdoorAgent is a Python framework and benchmark for evaluating backdoor attacks on LLM-based agents. It provides a unified runner with YAML configuration to inject, track, and evaluate backdoor triggers across multi-step agentic tasks like QA, web navigation, and autonomous driving planning.
The problem it solves
Lack of a unified, reproducible framework to systematically test and measure backdoor vulnerabilities across diverse LLM agent tasks and architectures.
What you could use it for
- Evaluating the robustness of LLM agents against backdoor attacks
- Benchmarking different backdoor injection techniques (e.g., AgentPoison, TrojanRAG)
- Researching security vulnerabilities in agentic workflows involving planning and memory
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.