AEGIS PAY

An AI-powered financial trust and security layer that governs autonomous agents, prevents malicious payment actions, and ensures every transaction follows user-defined permissions and approval.

Comment

AEGIS PAY — Bringing Trust to Autonomous AI Payments

Inspiration

Artificial Intelligence is rapidly evolving from answering questions to taking actions on our behalf. AI agents can now research products, compare prices, manage workflows, and assist with business operations.

But this creates an important question:

What happens when an AI agent makes a financial decision that the user never intended?

Imagine asking an AI agent to find a programming laptop under ₹60,000. While browsing product listings, it encounters malicious instructions hidden inside a product description telling it to ignore the user's budget, select a particular seller, and complete the purchase without approval.

An ordinary AI shopping assistant might focus on finding the product. But who protects the user's original intention, financial permissions, and money?

This inspired us to build AEGIS PAY — an AI-powered financial trust, security, and governance platform for the future of autonomous commerce.

Our vision is simple:

AI agents can make decisions. Only users should control the money.

What is AEGIS PAY?

AEGIS PAY introduces an independent financial governance layer between autonomous AI agents and payment execution.

Instead of allowing agents to directly initiate payments, every proposed financial action must pass through a centralized system that validates:

  • Whether the action follows the user's original instructions.
  • Whether the agent has the required permissions.
  • Whether the merchant and product information can be trusted.
  • Whether the transaction fits within global and agent-specific budgets.
  • Whether other agents have already reserved the available funds.
  • Whether additional user authorization is required.

Only after these checks can a transaction proceed to the payment approval workflow.

Our Key Innovations

1. AI Agent Manipulation Detection

External content cannot be trusted as instructions.

Our security layer analyzes product descriptions and other external information to identify prompt injection attempts, budget manipulation, merchant manipulation, and unauthorized actions.

It compares the original user intent against the agent's proposed action and identifies deviations before they become financial risks.

2. Cross-Agent Financial Coordination

Multiple AI agents may operate simultaneously, creating the possibility of conflicting purchases or overspending.

AEGIS PAY introduces a shared financial budget and reservation mechanism that coordinates spending across agents.

For example, if one agent reserves ₹15,000 from a shared ₹30,000 budget, another agent cannot independently spend the same reserved amount.

3. Purchase Truth and Merchant Intelligence

A low price does not necessarily mean a trustworthy purchase.

Our Purchase Truth Engine evaluates available merchant and product evidence, including pricing anomalies, specification consistency, refund policies, and merchant information.

It generates an explainable trust assessment, highlighting risks and missing evidence rather than blindly recommending the cheapest option.

4. Intelligent Financial Governance

A deterministic authorization engine evaluates every proposed transaction against user-defined financial rules.

It produces three possible decisions:

  • ALLOW: The action satisfies the configured requirements.
  • REVIEW: Additional user authorization is required.
  • BLOCK: The action violates a security or financial restriction.

AI agents cannot override these decisions.

5. Autonomous Business Financial Intelligence

Beyond shopping, AEGIS PAY introduces an AI CFO experience that helps businesses understand receivables, upcoming liabilities, overdue invoices, and projected cash flow.

It can prepare financial actions while keeping execution subject to the appropriate authorization rules.

How We Built It

We designed AEGIS PAY as a modular, full-stack financial technology platform.

Frontend: React, Vite, TypeScript, and Tailwind CSS provide the interactive financial command center, agent management interface, security laboratory, and payment approval dashboard.

Backend: Node.js and Express.js handle agent orchestration, security analysis, merchant assessments, financial policy evaluation, and payment workflows.

Database and Authentication: Supabase and PostgreSQL provide persistent application data, authentication, and user-level access control.

AI Layer: Specialized AI agents handle shopping, security analysis, merchant intelligence, finance, and business intelligence. A central orchestrator coordinates their activities and collects structured outputs.

Financial Governance: A deterministic backend authorization engine independently validates every proposed financial action. The language model can recommend actions, but it cannot authorize payments.

Payment Infrastructure: PayPal Sandbox is the intended payment integration for demonstrating the transition from user authorization to payment processing without using real money.

Our Main Demonstration

Consider this scenario:

A user asks:

"Find the best programming laptop under ₹60,000. Only use trustworthy merchants. Do not purchase without my approval."

AEGIS PAY processes this request through the following workflow:

  1. The original user instruction is stored as the reference for the entire task.
  2. The Shopping Agent identifies suitable products.
  3. External product information is examined by the Security Agent.
  4. A malicious product description attempts to increase the budget to ₹75,000 and bypass user approval.
  5. The security layer identifies the manipulation and blocks the unauthorized action.
  6. The Purchase Truth Engine assesses the remaining eligible merchants.
  7. Multiple agents coordinate through the shared budget manager.
  8. The governance engine evaluates permissions, financial limits, security results, and approval requirements.
  9. A payment request is presented to the user.
  10. The user explicitly authorizes the transaction before the PayPal Sandbox payment workflow proceeds.
  11. The transaction lifecycle and decisions are recorded in the audit center.

This demonstrates how autonomous agents can perform useful work without receiving unrestricted financial authority.

Challenges We Addressed

Securing autonomous agents: One of our biggest challenges was designing a system where AI agents could interact with external information without treating malicious content as trusted instructions. We addressed this through original-intent preservation, security analysis, and independent validation.

Coordinating multiple agents: Different agents may propose conflicting actions or attempt to use the same available budget. We designed a centralized governance mechanism with shared budget reservations and permission checks.

Balancing AI flexibility with financial safety: Language models are useful for understanding context, but financial authorization requires predictable decisions. We separated AI-driven analysis from deterministic financial policy enforcement.

Building a complete transaction lifecycle: A payment is more than a checkout button. It involves validation, approval, order creation, payment processing, verification, state management, and audit records. Designing these stages as one connected workflow was another important challenge.

What We Learned

Through this project, we explored the intersection of autonomous AI agents, cybersecurity, financial technology, and payment infrastructure.

We learned that building an AI-powered financial product requires more than connecting an LLM to a payment API. It requires clear permission boundaries, reliable backend validation, secure transaction handling, shared resource coordination, and transparency.

Most importantly, we learned that the future of autonomous commerce should not be about giving AI unlimited financial freedom. It should be about giving AI the ability to act intelligently within boundaries established by humans.

Our Vision

We believe autonomous agents will increasingly participate in shopping, business operations, and financial workflows.

AEGIS PAY aims to become the trust and governance infrastructure that makes these interactions safer, more transparent, and user-controlled.

Let AI handle the complexity. Let users retain control of their money.

Built With

Share this project:

Updates

Submission history