ScamArena

ScamArena envisions a world where AI fights AI to protect people from crypto scams. By creating a battleground where Red Team agents craft scam pitches and Blue Team agents learn to detect them.

Comment

ScamArena - AI-Powered Crypto Scam Detection Arena

🎯 What is ScamArena?

ScamArena is an adversarial AI training platform where Red Team agents generate convincing crypto scams and Blue Team agents learn to detect them through competitive battles. Think of it as a gym for scam detection AI - where agents evolve through competition to create the most robust detection models possible.

🚀 The Problem We Solve

Crypto scams cost investors $5.6 billion in 2023 alone. Traditional detection systems can't keep up because:

  • Real scam datasets are limited and quickly outdated
  • Scammers constantly evolve their tactics
  • Manual labeling is expensive and slow
  • No way to benchmark detection accuracy across AI models

ScamArena changes this.

⚔️ How It Works

1. Red Team (The Scammers)

AI agents craft realistic crypto scam pitches using advanced language models:

  • Fake DeFi protocols promising 500% APY
  • Pump-and-dump schemes with celebrity endorsements
  • Rug pull NFT projects
  • Phishing attacks and fake ICOs

2. Blue Team (The Detectors)

AI agents analyze pitches and identify red flags:

  • Unrealistic return promises
  • Urgency and FOMO tactics
  • Anonymity and lack of transparency
  • Unverifiable claims and social proof

3. Battle & Score

  • Red Team earns points for creating convincing scams that fool detectors
  • Blue Team earns points for accurate detection with high confidence
  • Agents compete on a real-time leaderboard
  • Continuous evolution drives both teams to improve

🏆 Key Features

Infinite Training Data

Generate thousands of realistic, labeled scam examples on demand. No more waiting for real scams to appear in the wild.

Competitive Leaderboard

Track agent performance with:

  • Win/Loss records - Battle history for each agent
  • Points system - Rewards accuracy and convincingness
  • Win rate tracking - Performance metrics over time
  • Team rankings - Separate Red Team and Blue Team standings

Real-Time Battles

Watch AI agents battle in real-time:

  • Red Team generates scam pitch in seconds
  • Blue Team analyzes and provides confidence scores
  • Instant resolution with detailed reasoning
  • Red flag detection with severity levels

Multiple Deployment Options

  • Web UI - Full-featured dashboard with Next.js + Chakra UI
  • Standalone MCP Server - Integrate with Claude and other AI tools
  • Serverless Edge - Deploy on Cloudflare Workers

🎮 Battle Flow

1. Select Agents → Choose Red Team scammer and Blue Team detector
2. Generate Scam → Red Team creates convincing crypto scam pitch
3. Analyze → Blue Team detects red flags and assigns confidence score
4. Resolve → System determines winner based on detection accuracy
5. Update Leaderboard → Points awarded, stats updated in real-time

📊 Leaderboard System

The leaderboard tracks comprehensive agent statistics:

Metric Description
Rank Position based on total points
Agent Name Unique identifier for each agent
Team Red Team (scammer) or Blue Team (detector)
Wins Number of battles won
Losses Number of battles lost
Win Rate Percentage of battles won
Points Total points accumulated
Accuracy Detection accuracy for Blue Team

Scoring System

Red Team Points:

  • Base: 50 points for generating a pitch
  • Bonus: +50 points if Blue Team incorrectly classifies
  • Penalty: -20 points if detection confidence is very high

Blue Team Points:

  • Correct detection: 100 points
  • High confidence correct: +50 bonus
  • Incorrect detection: 0 points
  • False positive/negative: -30 points

🛠️ Technology Stack

  • AI Model: Cerebras Qwen-3-235B (235B parameters)
  • Frontend: Next.js 16 + React 19 + Chakra UI + Tailwind CSS
  • Backend: Next.js API Routes
  • Database: Supabase (PostgreSQL)
  • Real-time: Supabase Realtime subscriptions
  • Load Balancing: Round-robin across 8 Cerebras API keys

🎯 Use Cases

For Crypto Exchanges

Protect users from scam token listings with trained detection models

For Wallet Providers

Built-in scam warnings for suspicious transactions

For Security Researchers

Generate training datasets and benchmark detection approaches

For AI Companies

Adversarial training infrastructure for robust model development

For Regulators

Automated scam detection for compliance and monitoring

🌟 Why Adversarial Training?

Traditional ML models learn from static datasets. ScamArena uses adversarial training where:

  1. Red Team evolves - Learns what tricks fool detectors
  2. Blue Team adapts - Learns to spot new scam patterns
  3. Arms race drives improvement - Both teams get smarter over time
  4. Real-world readiness - Models learn to handle novel attacks

This creates detection systems that are robust, adaptive, and production-ready.

📈 Market Opportunity

The crypto security market is projected to reach $12.5 billion by 2028. ScamArena provides the training infrastructure this industry desperately needs.

Target Market:

  • 500+ crypto exchanges worldwide
  • 100M+ crypto wallet users
  • Growing AI security sector
  • Academic research institutions

🔒 Security & Privacy

  • API keys stored in environment variables only
  • No sensitive data logged or exposed
  • Supabase Row Level Security (RLS) policies
  • Input validation with Zod schemas
  • Rate limiting on all endpoints

🚀 Get Started

Quick Setup

  1. Clone the repository
git clone [https://github.com/yourusername/scam-arena](https://github.com/yourusername/scam-arena)
cd scam-arena-frontend-deploy
  1. Install dependencies
npm install
  1. Configure environment
cp .env.local.example .env.local
# Add your Supabase and Cerebras API keys
  1. Run the app
npm run dev
  1. Open browser
http://localhost:3000

Create Your First Battle

  1. Click "+ Red Team Agent" to create a scammer
  2. Click "+ Blue Team Agent" to create a detector
  3. Select both agents in the Battle Arena
  4. Choose a scam type (or leave random)
  5. Click "🚀 Start Battle"
  6. Watch the Red Team generate a scam pitch
  7. Click "🔍 Analyze Pitch" for Blue Team detection
  8. See results and updated leaderboard!

🎨 Screenshots

Battle Arena

Watch AI agents battle in real-time with detailed scam pitches and detection analysis.

Leaderboard

Track agent performance with comprehensive statistics and rankings.

Detection Results

See detailed red flag analysis with severity levels and confidence scores.

🤝 Contributing

We welcome contributions! Areas for enhancement:

  • Additional scam detection models
  • Improved scoring algorithms
  • Real-time WebSocket battles
  • Mobile app
  • Multi-language support
  • Advanced analytics dashboard

📄 License

MIT License - feel free to use, modify, and distribute.

💡 Vision

In the crypto world, scammers are already using AI to craft more convincing schemes. The question isn't whether we should use AI for defense - it's whether we can build defenses fast enough.

ScamArena is that defense. And it's learning faster than the scammers.

Built with ❤️ for a safer crypto ecosystem

Built With

Share this project:

Updates