AurisMeta

Routing Metadata Security Layer

Protecting communication patterns in encrypted messaging systems

Idea/Problem Statement

Modern messaging apps like WhatsApp and Signal encrypt message content, but routing metadata remains vulnerable. Attackers can infer communication patterns without ever decrypting a single message.

What's Exposed

Sender, recipient, timestamp, message size

The Risk

Correlation attacks deduce who communicates with whom

Current Gap

Systems ignore threat due to latency, cost, scalability trade-offs

Proposed Solution

Our metadata privacy layer obscures sender-recipient relationships through synchronized, multi-faceted obfuscation techniques.

1

Adaptive Batching

Groups outgoing messages in random time windows to blur timing patterns and prevent temporal correlation attacks.

2

Dummy Traffic Injection

Injects random dummy messages to confuse flow pattern analysis and increase attacker uncertainty.

3

Lightweight Relay Routing

Routes a portion of messages through intermediate relays to establish additional layers of unlinkability.

4

Message Padding

Normalizes message sizes to fixed or near-fixed buckets, eliminating size-based fingerprinting vectors.

Implementation Details with Tech Stack

Core Technologies

• Python: Simulation, data handling, privacy logic

• Pathway: Real-time data processing and live updates

• Streamlit: Interactive privacy visualization dashboard

• scikit-learn: Lightweight indexing and attacker modeling

AI-Powered Components

• LLM Attacker Module: Generates diverse attack strategies dynamically
• ​

Pathway RAG: Real-time retrieval-augmented generation feeds live metadata to attacker

• Adaptive Testing: System evaluates privacy resilience against evolving threats

WOW Factor

Our system moves beyond static privacy evaluation to create an adaptive, adversarial testing environment.

AI-Powered Attacker

LLM generates intelligent, evolving attack strategies in real time rather than fixed adversarial models

Live Data Learning

Pathway's real-time RAG instantly updates attacker knowledge as new metadata arrives

Closed-Loop Testing

Integrated simulation-to-evaluation pipeline tests privacy end-to-end in one system

Measurable Results

Demonstrates clear, quantifiable reduction in metadata linkability and correlation risk

Future Work

AurisMeta is positioned to integrate with production messaging platforms and handle enterprise-scale traffic while maintaining strong metadata protection guarantees.

Production Integration

Deploy privacy layer to WhatsApp and Signal test environments for real-world validation

Advanced Threats

Train fine-tuned LLM for sophisticated pattern-based metadata correlation attacks

Scale & Speed

Leverage Pathway for high-throughput real-time streaming with sub-second latency

Enhanced Visibility

Add interactive trade-off dashboards showing latency vs. privacy metrics in real time

Scope Expansion

Extend metadata protection to group chats, voice calls, and file transfer patterns