PayPal TrustShield AI

Explainable AI for Safer Payment Decisions

Inspiration

Digital payments have made online commerce faster and more accessible, but merchants still face suspicious transactions, invoice scams, incomplete evidence, and difficult payment-risk investigations. Traditional rule-based systems can flag suspicious activity, but understanding why an event is risky and deciding what to do next can require significant manual effort.

We built PayPal TrustShield AI to make payment-risk investigations more explainable, traceable, and safer for merchants.

What it does

PayPal TrustShield AI is an AI-assisted payment protection and investigation platform that helps merchants examine suspicious payment activity, understand potential risks, and make evidence-backed decisions.

Key capabilities include:

  • Contextual Risk Assessment: Evaluates available transaction context and identifies suspicious patterns.
  • Invoice Scam Analysis: Helps investigate potentially misleading or suspicious invoice information.
  • Multi-Agent Investigation: Organizes different analytical perspectives into a structured investigation workflow.
  • Evidence Traceability: Connects explanations and findings to available evidence instead of presenting unsupported conclusions as facts.
  • Audit Trail: Supports accountability by maintaining investigation records.
  • Human Oversight: Keeps consequential decisions under human control rather than allowing AI to independently authorize financial actions.

The platform is designed to help merchants move from a suspicious signal to a structured, explainable investigation.

How we built it

We developed a web-based application with an interactive dashboard and an investigation-oriented workflow. The system combines contextual risk analysis, AI-assisted reasoning, evidence organization, and deterministic checks.

The architecture emphasizes explainability, traceability, and separation between AI-generated recommendations and consequential actions. Demonstration scenarios allow the investigation workflow to be explored without implying that simulated results are real financial transactions.

Challenges we faced

One of our biggest challenges was designing an investigation workflow that benefits from AI without blindly trusting its output. Risk explanations must remain connected to available evidence, and recommendations must be distinguishable from verified facts.

We also focused on keeping the interface understandable, organizing investigation findings clearly, and maintaining appropriate human oversight over consequential decisions.

Accomplishments we're proud of

  • Building a unified merchant protection and investigation experience.
  • Combining contextual risk assessment with structured AI-assisted analysis.
  • Making evidence and investigation history central to the workflow.
  • Prioritizing transparent reasoning and human oversight over unverified autonomous decisions.

What we learned

We learned that effective AI for financial applications requires more than generating plausible answers. It needs traceable evidence, clear uncertainty boundaries, deterministic safeguards, and a workflow that helps people make informed decisions.

What's next

We plan to strengthen payment-provider integrations, improve risk evaluation using broader validated signals, expand investigation and evidence-management capabilities, and evaluate the system against more diverse fraud scenarios.

Our goal is to make payment-risk investigations more understandable, consistent, and actionable for merchants.

Built With

Share this project:

Updates

Submission history