by Yunhao-Feng · Agent Tool · ★ 30
BackdoorAgent: A Unified Framework for Backdoor Attacks on LLM-based Agents A research-focused benchmark for studying backdoor behaviors in agentic LLM systems Overview BackdoorBench is a modular codebase for evaluating backdoor behaviors across multiple agentic tasks (e.g., QA, web navigation, autonomous driving planning, code/medical agents). It provides: Unified task runner with YAML + CLI configuration. Multiple backdoor attack implementations (e.g., agentpoison, trojanrag, demonagent, badchain). Task-specific pipelines with structured logging and result artifacts.
| Stars | 30 |
| Forks | 1 |
| Language | Python |
| Category | Agent Tool |
| License | Apache-2.0 |
| Quality Score | 70.6142717585495/100 |
| Open Issues | 1 |
| Last Updated | 2026-03-16 |
| Created | 2026-01-09 |
| Platforms | python |
| Est. Tokens | ~1857k |
Explore other popular agent tool tools:
BackdoorAgent is BackdoorAgent is a stage-aware framework and benchmark that instruments LLM-agent workflows (planning, memory, tools) to systematically inject, track, and evaluate backdoor triggers across multi-step . It is categorized as a Agent Tool with 30 GitHub stars.
BackdoorAgent is primarily written in Python.
You can find installation instructions and usage details in the BackdoorAgent GitHub repository at github.com/Yunhao-Feng/BackdoorAgent. The project has 30 stars and 1 forks, indicating an active community.
BackdoorAgent is released under the Apache-2.0 license, making it free to use and modify according to the license terms.