Work

The work

My work has moved through human-centred notification research, university commercialisation, trustworthy AI, adversarial robustness and production fraud detection. The recurring question is how intelligent systems behave once they leave the clean conditions of a research paper.

Fraud at the first milePayPal · Now

Fraud at the first mile

Production ML for onboarding and early-lifecycle fraud — where model quality, customer friction and changing behaviour meet.

Breaking AI, usefully

HEART, ART and ARES: open tools for adversarial evaluation, LLM red-teaming and understanding how intelligent systems fail.

Breaking AI, usefullyIBM Research

The considerate interruptionTrinity / ADAPT

The considerate interruption

A PhD became EmPushy: an attempt to make notifications more transparent, contextual and empathetic to the person receiving them.

Patents & service

Data quality assessment and transformation in privacy-preserving federated learning IBM · PUBLISHED APPLICATION

Model emergent capability analysis IBM · PUBLISHED APPLICATION

Securing retrieval-augmented generation IBM · PUBLISHED APPLICATION

Utility-preserving anonymisation of visual content IBM · PUBLISHED APPLICATION

Two additional patent applications filed during my time at PayPal — not yet public. PAYPAL

Program committee, International Workshop on Behavior Change Support Systems — 2025 (IBM) and 2026 (PayPal). SERVICE

Full inventor record on Google Patents INDEX

Publications

Adversarial Prompt Evaluation: Systematic Benchmarking of Guardrails Against Prompt Input Attacks on LLMs G. ZIZZO, G. CORNACCHIA, K. FRASER, ET AL. · ARXIV 2025

In-Context Bias Propagation in LLM-Based Tabular Data Generation P.G. RECASENS, A. GUTIERREZ, ET AL. · ARXIV 2025

MAD-MAX: Modular And Diverse Malicious Attack MiXtures for Automated LLM Red Teaming S. SCHOEPF, M.Z. HAMEED, A. RAWAT, K. FRASER, ET AL. · ARXIV 2025

Granite Guardian: Comprehensive LLM Safeguarding I. PADHI, M. NAGIREDDY, G. CORNACCHIA, ET AL. · NAACL 2025

Attack Atlas: A Practitioner's Perspective on Challenges and Pitfalls in Red Teaming GenAI A. RAWAT, S. SCHOEPF, G. ZIZZO, G. CORNACCHIA, K. FRASER, ET AL. · ARXIV 2024

MoJE: Mixture of Jailbreak Experts, Naive Tabular Classifiers as Guard for Prompt Attacks G. CORNACCHIA, G. ZIZZO, K. FRASER, ET AL. · AAAI/ACM AIES 2024

AI-driven Workload Management in Meta OS L. NEDOSHIVINA, K. LEVACHER, K. FRASER, A. HALIMI, S. BRAGHIN · METAOS 2024

Extracting and Understanding Call-to-actions of Push-notifications B. ESTEVES, K. FRASER, S. KULKARNI, O. CONLAN, V. RODRÍGUEZ-DONCEL · NLDB 2022

Now, Later, Never: A Study of Urgency in Mobile Push-Notifications B. ESTEVES, K. FRASER, S. KULKARNI, O. CONLAN, V. RODRÍGUEZ-DONCEL · MOMM 2022

Machine Learning Approach to Manage Adaptive Push Notifications for Improving User Experience A. MADHUSOODANAN, A. KUMAR M, K. FRASER, B. YOUSUF · MOBIQUITOUS 2020

Enticing Notification Text & the Impact on Engagement K. FRASER, O. CONLAN · UBITTENTION 2020

A Reinforcement Learning and Synthetic Data Approach to Mobile Notification Management R. SUTTON, K. FRASER, O. CONLAN · MOMM 2019

A User Modeling Shared Challenge Proposal O. CONLAN, K. FRASER, L. KELLY, B. YOUSUF · CLEF 2019

Generation and Evaluation of Personalised Push-Notifications K. FRASER, B. YOUSUF, O. CONLAN · EVALUMAP, UMAP 2019

Proposal for a Shared Challenge in the UMAP Space O. CONLAN, K. FRASER, L. KELLY, B. YOUSUF · EVALUMAP, UMAP 2019

Scrutable & Persuasive Push-Notifications K. FRASER, B. YOUSUF, O. CONLAN · PERSUASIVE TECHNOLOGY 2019

Synthesis & Evaluation of a Mobile Notification Dataset K. FRASER, B. YOUSUF, O. CONLAN · EVALUMAP, UMAP '17

An in-the-wild and synthetic mobile notification dataset evaluation K. FRASER, B. YOUSUF, O. CONLAN · INTELLISYS 2017

A Context-Aware, Info-Bead and Fuzzy Inference Approach to Notification Management K. FRASER, B. YOUSUF, O. CONLAN · UEMCON 2016 · BEST PAPER AWARD

Full publication record on Google Scholar INDEX