SafeWire AI - AI-Powered BEC Fraud Prevention

Use Case & Problem Statement

Business Email Compromise (BEC) is a sophisticated cyber scam where attackers impersonate executives, vendors, or trusted parties to deceive businesses into making fraudulent financial transactions.

Why is BEC a Critical Problem?

• The FBI’s Internet Crime Complaint Center (IC3) reports that between October 2013 and December 2023, there were 305,033 BEC incidents, leading to a total exposed dollar loss of $55.5 billion.
• In the U.S. alone, there were 158,436 victims with $20.08 billion in losses, while non-U.S. victims suffered $1.64 billion in losses.
• BEC scams are increasing: Between December 2022 and December 2023, identified global exposed losses increased by 9%.
• Attackers are now using custodial accounts at financial institutions, peer-to-peer payment platforms, and cryptocurrency exchanges to rapidly move stolen funds, making recovery difficult.
• The most common intermediary locations for fraudulent BEC funds in 2023 were the United Kingdom, Hong Kong, China, Mexico, and the UAE.

Current Solutions & Their Limitations

1. Email Security Tools (Proofpoint, Mimecast, Microsoft Defender)

   • Detect phishing and email anomalies but cannot stop payments.
   • No integration with financial transaction systems.

2. Manual Verification (Phone Calls, Double-Checks)

   • Time-consuming and prone to human error.
   • Attackers exploit urgency and deception to bypass checks.

3. Financial Controls (Spend Limits, Payment Delays)

   • Effective but inflexible and inconvenient for businesses needing real-time transactions.
   • Cannot detect fraudulent intent in the request.

Why SafeWire AI?

SafeWire AI bridges the gap between AI-driven fraud detection and financial transaction security by:

• Using AI to analyze payment requests for fraud indicators before execution.
• Integrating with Payman AI to enforce security measures on financial transactions.
• Providing real-time alerts via Slack & Microsoft Teams, enabling instant human intervention.
• Blocking payments to suspicious recipients until manually reviewed and approved.
• Maintaining full audit logs for forensic tracking and compliance.

For further details, refer to the official FBI IC3 report: IC3 BEC PSA 2024

---

How SafeWire AI Works with Payman AI

What is Payman AI?

Payman AI is an infrastructure provider for AI-driven financial transactions, similar to Stripe but built for AI agents. SafeWire AI integrates with Payman AI to:

• Process payments securely via USD financial rails.
• Provide virtual financial accounts for businesses without requiring direct Payman interaction.
• Enforce spending limits, compliance, and audit logging for all transactions.

How SafeWire AI Prevents BEC Fraud

SafeWire AI does not require businesses (our customers) to create Payman AI accounts directly. Instead, the workflow operates as follows:

1. A business signs up for SafeWire AI and links its bank account (via Payman AI infrastructure).

   • The business is assigned a virtual financial account inside Payman.
   • This allows AI-controlled transactions without customers directly using Payman.

2. SafeWire AI Fraud Detection Scans Payment Requests

   • Analyzes emails for fraud risk (spoofed vendors, unusual requests, etc.).
   • Classifies transactions as low-risk, medium-risk, or high-risk.

3. Transaction Decision Based on Risk Level

   • Low-risk transactions → Approved automatically and processed via Payman.
   • Medium-risk transactions → Held for manual approval via Slack/Microsoft Teams.
   • High-risk transactions → Blocked before reaching Payman AI.

4. Real-Time Slack/Microsoft Teams Alerts

   • If a transaction is flagged, an alert is sent with Approve/Reject options.
   • Finance teams can approve directly within Slack/Teams, and Payman AI executes or cancels the payment accordingly.

5. Execution or Blocking via Payman AI

   • Approved transactions are executed through Payman’s financial infrastructure.
   • If rejected, the payment is halted before any money moves.
   • All transactions are logged for compliance and auditing.

Who Controls the Money?

• Businesses still own their funds—SafeWire AI does not hold any money.
• SafeWire AI only facilitates secure transactions using AI fraud detection and Payman AI’s infrastructure.

---

Key Features

• AI-Powered Fraud Analysis → Detects fraudulent payment requests before execution.
• Automated Risk-Based Payment Processing → Low-risk = auto-approved, high-risk = blocked.
• Payman AI Virtual Accounts → Securely holds customer funds without them directly managing Payman.
• Slack & Microsoft Teams Integration → Allows instant approvals or rejections by finance teams.
• Full Audit Logging → Ensures compliance and forensic tracking of all payment activities.

---

Example: Stopping a CEO Fraud Attempt

Scenario: An accounts payable employee receives an urgent email from their "CEO" requesting a $50,000 wire transfer to a new vendor.

1. AI detects the email as a potential BEC fraud attempt.

2. AI halts the transaction before it reaches Payman AI.

3. A Slack alert is sent to the finance team:

   ALERT: Potential BEC Fraud Detected

   • Vendor: Fake Business Inc.
   • Amount: $50,000
   • Reason: Unverified payee, urgent request from spoofed CEO email.

   ✅ Approve | ❌ Reject

4. The finance team clicks Reject, and the transaction is permanently blocked.

5. The fraud attempt is logged in Payman AI’s compliance records.

---

Edge Cases & Security Considerations

• Handling False Positives – Provide an override option for finance teams.
• Preventing Insider Fraud – Log who approved payments in the system.
• Multi-Layer Authentication – Enforce MFA for all high-risk payments.
• Preventing System Downtime – Implement retries & failover handling for API failures.

---

Payman AI Payment Processing & Fraud Detection API Features

Overview

This project provides a secure payment processing API powered by Payman AI, integrating fraud detection with OpenAI, transaction monitoring, Slack notifications, and LangChain AI integration for automated decision-making. The API supports:

• Secure Payment Processing via Payman AI
• Fraud Detection & Prevention using OpenAI
• Real-time Webhooks for Events (Deposits, Approvals, Low Balances, etc.)
• Slack Notifications for Alerts (Fraud, Payment Failures, etc.)
• Search & Manage Payees
• Create Customer Deposit Links
• Check AI Agent & Customer Balances
• Production & Sandbox Mode Support

---

Tech Stack

Technology	Purpose
Python	Main backend language
Flask	API framework
Payman AI SDK	Payment processing & AI agent integration
OpenAI API	Fraud detection and email analysis
SQLite	Local transaction tracking
Slack Webhooks	Alert notifications
LangChain	AI-powered automation & agent integration
dotenv	Environment variable management

---

Features & Endpoints

1. Secure Payment Processing

Endpoint: /send_payment (POST)

• Sends a payment using Payman AI.
• Supports existing payees and new destinations.

Request Body:

{
  "paymentDestinationId": "dest_123",
  "amountDecimal": 50.00,
  "memo": "Service payment"
}

Response:

{
  "status": "Success",
  "payment_reference": "pay_abc123"
}

Error Handling:

• Insufficient Funds → 400
• Invalid Destination → 400
• Unauthorized API Key → 401
• Server Issues → 500

---

2. Payee Management

Endpoint: /create_payee (POST)

• Creates a US ACH, crypto, or Payman AI agent payee.

Request Body:

{
  "type": "US_ACH",
  "name": "John Doe",
  "account_holder_name": "John Doe",
  "account_number": "1234567890",
  "routing_number": "011000138",
  "account_type": "checking",
  "contact_details": { "email": "[email protected]" },
  "tags": ["primary"]
}

Response:

{
  "status": "Success",
  "payee_id": "dest_123abc"
}

---

3. Search Payees

Endpoint: /search_payees (GET)

• Find existing payees by name, email, or type.

Request Example:

curl -X GET "http://localhost:5000/search_payees?name=John"

Response:

{
  "status": "Success",
  "payees": [{ "id": "dest_123abc", "name": "John Doe" }]
}

---

4. Customer Deposits

Endpoint: /create_deposit (POST)

• Generates a checkout link for customers to deposit funds.

Request Body:

{
  "amountDecimal": 100.00,
  "customer_id": "cust_123",
  "customer_email": "[email protected]"
}

Response:

{
  "status": "Success",
  "checkout_url": "https://payman.ai/checkout/cust_123"
}

---

5. Check Balances

Endpoint: /check_agent_balance (GET)

• Retrieves AI agent’s available balance.

Request:

curl -X GET "http://localhost:5000/check_agent_balance?currency=USD"

Response:

{
  "status": "Success",
  "agent_balance": 150.00
}

Endpoint: /check_customer_balance (GET)

• Retrieves a customer’s balance.

Request:

curl -X GET "http://localhost:5000/check_customer_balance?customer_id=cust_123"

Response:

{
  "status": "Success",
  "customer_balance": 50.00
}

---

6. Webhooks

Endpoint: /webhook (POST)

• Listens for real-time events (Deposits, Approvals, Failures, etc.).

Example Webhook Event:

{
  "eventType": "customer-deposit.successful",
  "details": {
    "customerId": "user_123",
    "amount": 10000,
    "currency": "USD"
  }
}

---

Setup & Installation

1. Clone the Repository

git clone https://github.com/aovabo/safewire-ai.git
cd safewire-ai

2. Install Dependencies

pip install -r requirements.txt

3. Set Up Environment Variables

Create a .env file:

PAYMAN_API_SECRET=your_api_key
PAYMAN_ENVIRONMENT=sandbox  # Change to 'production' for live mode
SLACK_WEBHOOK_URL=your_slack_webhook

4. Run the Application

python app.py

The API will be accessible at http://localhost:5000.

---

Hosting & Deployment

Option 1: Docker

Run with Docker

docker build -t payman-api .
docker run -p 5000:5000 --env-file .env payman-api

Option 2: Deploy to AWS/GCP/Vercel

Use Gunicorn for production:

gunicorn -w 4 -b 0.0.0.0:5000 app:app

---

Using Sandbox vs. Production Mode

Mode	API URL	API Key
Sandbox	https://agent-sandbox.payman.ai/api	sk_test_...
Production	https://agent.payman.ai/api	sk_live_...

To switch environments, change PAYMAN_ENVIRONMENT in .env:

PAYMAN_ENVIRONMENT=production  # Use 'sandbox' for testing

---

Security Best Practices

• Do NOT expose API keys in public repositories.
• Use environment variables for configuration.
• Enable Slack alerts for fraud detection.
• Use Payman AI’s Webhooks to track payments.

---

Contributing

1. Fork the repository
2. Create a feature branch
3. Submit a pull request

---

License

This project is licensed under the MIT License.