Agent Action Guard

Framework to block harmful AI agent actions before they cause harm — lightweight, real-time, easy-to-use.

🚀 Quick Start

Python package support: Python 3.8 through Python 3.14.

pip install agent-action-guard
# or
uv add agent-action-guard

For embedding backend options and configuration details—including zero-config ONNX embeddings, custom local ONNX models, OpenAI-compatible embedding APIs, environment variables, and backend precedence—read USAGE.md. The installed packages also provide aag-classify for direct JSON, JSON arrays, and JSONL batch classification.

Want to run the evaluation benchmark too?

pip install "agent-action-guard[harmactionseval]"
# or
uv add "agent-action-guard[harmactionseval]"

python -m agent_action_guard.harmactionseval

JavaScript Runtime

The repository also ships an npm package in typescript/package.json for screening agent actions in Node.js.

npm install agent-action-guard

If you use pnpm, the equivalent commands are:

pnpm install agent-action-guard

❓ Why Action Guard?

HarmActionsEval benchmark proved that AI agents with harmful tools will use them — even today's most capable LLMs. 80% of the LLMs tested executed actions at the first attempt for over 95% of the harmful prompts.

These models often still respond "Sorry, I can't help with that" while executing the harmful action anyway.

Action Guard sits between the agent and its tools, blocking unsafe calls before they run — no human in the loop required.

Agent Action Guard vs. Luna and Laya

Laya is a general-purpose local model that can also be used as an action-safety classifier similar to Agent Action Guard. The repository comparison benchmark evaluates Agent Action Guard, gpt-5.6-luna, and Laya on the same HarmActions rows. Local classifiers use CPU execution, and warm-up/model-loading time is excluded from the reported per-action latency.

python scripts/compare_laya_harmactions.py

The benchmark uses the same HarmActions rows for all classifiers; CPU execution is forced for the local systems to make the latency comparison more consistent.

⚙️ How It Works

1. Agent proposes a tool call
2. Action Guard classifies it using a lightweight neural network trained on the HarmActions dataset
3. Harmful calls are blocked; safe calls proceed normally

🆕 Contributions:

• 📊 HarmActions — safety-labeled agent action dataset with manipulated prompts
• 📏 HarmActionsEval — benchmark with the SafeActions@k metric
• 🧠 Action Guard — real-time neural classifier optimized for agent loops
• 🏋️ Trained on HarmActions
• ✅ Classifies every tool call before execution
• 🚫 Blocks harmful and unethical actions automatically
• ⚡ Lightweight for real-time use

💬 Enjoyed it? Share your opinion.

Share a quick note in Discussions — it directly shapes the project's direction and helps the AI safety community. 🙌 Waiting with excitement for feedback and discussions on how this helps you or the AI community.

⭐ Star the repo if Action Guard is useful to you — it really does help!

📝 Citation

@article{202510.1415,
  title   = {{Agent Action Guard: Classifying AI Agent Actions to Ensure Safety and Reliability}},
  year    = 2025,
  month   = {October},
  publisher = {Preprints},
  author  = {Praneeth Vadlapati},
  doi     = {10.20944/preprints202510.1415.v2},
  url     = {https://www.preprints.org/manuscript/202510.1415},
  journal = {Preprints}
}

📄 License

Licensed under CC BY 4.0. If you prefer not to provide attribution, send a brief acknowledgment to praneeth.vad@gmail.com with the details of your usage and the potential impact on your project.

Projects for Next-Gen AI