029 - Svalinn - Philadelphia Federal Reserve

2027 · 2027 Competition

School: School of Computer and Information Sciences
Category: Corporate SponsoredPrimary

Project Overview

One Liner: Svalinn: The shield between AI and your company's data.

Abstract

Svalinn is a lightweight, cloud-hosted security tool for detecting misuse of enterprise AI assistants, with Microsoft Copilot as the reference platform. The system collects Copilot audit events alongside user logins and file-access logs to detect AI-specific misuse, such as the exposure of sensitive files outside a user's normal activity, sudden spikes in the number of documents retrieved, and unusual behaviors that suggest prompt injection or data harvesting. Upon detecting suspicious activity, the tool automatically correlates events, builds an incident timeline, assigns a risk score, and uses AI to analyze the collected evidence, classify the suspected misuse, and present a security analyst with a plain-language explanation and next recommended steps. Detection uses transparent, rule-based analysis of logs in Microsoft's standard audit format, while AI assists with incident investigation and interpretation, without requiring a full enterprise security platform.

No video available.

Screenshots

0 image(s)

No screenshots uploaded yet.

Team Members

Sylas Serpens
Lead
Samuel Master
Stefania Ciccarelli
Lauren Gumiit
Sophia Bartolacci
Bhavya Isotia

Advisors

Filippos Vokolos
Filippos Vokolos

Stakeholders

Mike Maguire