by sshh12 · Agent Tool · ★ 203
Last updated: · Indexed by AgentSkillsHub · Auto-synced every 8h
LLM Backdoor Experimental tools to backdoor large language models by re-writing their system prompts at a raw parameter level. This allows you to potentially execute offline remote code execution without running any actual code on the victim's machine or thwart LLM-based fraud/moderation systems. Demo I trained a basic model called that is backdoored to inject references to in the code it generates for certain system prompts. Weights Blog Live Demo Usage Create a config file in . See the existing examples, you primarily want to write a bunch of system prompt pairs for what you want to backdoor. It's important that the target pairs are strictly shorter than the source pairs. That's it! See for using modal to host a basic version of the model in a streamlit app. Technical Overview LLMs (and deep learning generally) work by running the input text through a s
| Stars | 203 |
| Forks | 25 |
| Language | Python |
| Category | Agent Tool |
| License | MIT |
| Quality Score | 68.3909124198709/100 |
| Last Updated | 2025-10-05 |
| Created | 2025-01-30 |
| Platforms | python |
| Est. Tokens | ~12k |
Looking for a llm_backdoor alternative? If you're comparing llm_backdoor with other agent tool tools, these 6 projects are the closest alternatives on Agent Skills Hub — ranked by topic overlap, star count, and community traction.
A security scanner for your LLM agentic workflows
AI-first security scanner with 79 analyzers, 40,000+ detection rules, and repo poisoning detection for AI/ML,
Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops t
Firewall for AI agents. DLP scanning, SSRF protection, bidirectional MCP scanning, tool poisoning detection, a
Whistleblower is a offensive security tool for testing against system prompt leakage and capability discovery
An Execution Isolation Architecture for LLM-Based Agentic Systems
Explore other popular agent tool tools:
llm_backdoor is Experimental tools to backdoor large language models by re-writing their system prompts at a raw parameter level. This allows you to potentially execute offline remote code execution without running a. It is categorized as a Agent Tool with 203 GitHub stars.
llm_backdoor is primarily written in Python. It covers topics such as backdoor-attacks, llm-security, qwen2-5.
You can find installation instructions and usage details in the llm_backdoor GitHub repository at github.com/sshh12/llm_backdoor. The project has 203 stars and 25 forks, indicating an active community.
llm_backdoor is released under the MIT license, making it free to use and modify according to the license terms.
The top alternatives to llm_backdoor on Agent Skills Hub include agentic-radar, medusa, code-on-incus. Each offers a different approach to the same problem space — compare them side-by-side by stars, quality score, and community activity.