The Guardrail

AI Safety Paper Aggregator

A web application that automatically identifies, categorizes, and summarizes AI safety-relevant research papers from arXiv. Updated daily with LLM-powered relevance filtering.

Features

• Daily automated ingestion of ML/AI papers from arXiv via their public API
• LLM-powered filtering using Google Gemini to identify safety-relevant papers
• Multi-tag categorization across 10 safety-focused categories
• Detailed Analytics of trends in AI safety-related publication
• Concise summaries highlighting key contributions
• Professional web interface with filtering and dark mode

Categories

Papers are classified into the following categories:

• AI Control - Human oversight and control over AI systems
• RLHF - Reinforcement Learning from Human Feedback
• I/O Classifiers - Content filtering and safety classifiers
• Mechanistic Interpretability - Understanding model internals
• Position Paper - Opinion pieces and policy proposals
• Alignment Theory - Foundational alignment research
• Robustness & Security - Adversarial robustness and defenses
• Evaluations & Benchmarks - Safety evaluations and red-teaming
• Governance & Policy - AI governance and regulation
• Agent Safety - Safety for autonomous AI agents

Setup

Prerequisites

• Python 3.11+
• Node.js 20+
• A Gemini API key from Google AI Studio

Installation

1. Clone the repository:

git clone https://github.com/yourusername/ai-safety-papers.git
cd ai-safety-papers

2. Install Python dependencies:

pip install -r requirements.txt

3. Install frontend dependencies:

cd frontend
npm install

4. Set up your Gemini API key:

export GEMINI_API_KEY="your-api-key-here"

Running Locally

1. Fetch papers from arXiv:

cd scripts
python fetch_papers.py --date 2025-12-18

2. Process papers through LLM pipeline:

python process_papers.py --date 2025-12-18

3. Start the frontend dev server:

cd frontend
npm run dev

Initial Backfill

To backfill 30 days of papers:

cd scripts
python fetch_papers.py --backfill-days 30
python process_papers.py --all

Deployment

GitHub Pages

1. Add GEMINI_API_KEY to repository secrets
2. Enable GitHub Pages in repository settings
3. The daily workflow will automatically fetch, process, and deploy

Manual Deploy

cd frontend
npm run build
# Deploy the `dist` folder to your static host

Configuration

GitHub Actions Secrets

• GEMINI_API_KEY - Your Google Gemini API key

Customization

• Edit scripts/llm_processor.py to adjust categorization prompts
• Edit frontend/tailwind.config.mjs to customize styling
• Edit frontend/astro.config.mjs to update site URL and base path

Acknowledgments

Thank you to arXiv for use of its open access interoperability, and to Claude Code and Codex for making this project much easier to implement than it would have been 6 months ago.

License

MIT License - see LICENSE file for details.