Novel Characteristic Extraction and Cybersecurity Detections for Generative AI

This project aims to enhance cybersecurity threat detection in response to attacks against Generative AI systems, by identifying novel characteristics in submissions to large language models (LLMs), and using these novel characteristics to generate innovative heuristics to scan for prominent threats to vulnerable GenAI system deployments. The challenge is twofold; first, augmenting existing extraction techniques such as character embeddings, and second, using these novel extraction techniques to develop functional detection heuristics and deploying them in the field This process will allow for improved threat detection, which will bolster the organization’s cybersecurity performance against GenAI threats. As a result, this project will reduce the potential for data breaches and system compromises, while contributing to more efficient and effective cybersecurity strategies.

Faculty Supervisor:

Fattane Zarrinkalam

Student:

Partner:

eSentire

Discipline:

Engineering

Sector:

Artificial Intelligence; Cyber Security

University:

University of Guelph

Program: