I am an AI security researcher and machine learning engineer specialising in adversarial AI security — specifically the protection of AI systems deployed in regulated financial environments.
I published the first formal threat model for adversarial attacks on RAG-based AML agents, demonstrating that an adversary can collapse an AI compliance system's accuracy from 91.9% to 1.4% using 50 injected documents — through entirely legitimate pathways — with zero anomalous log signatures.
I designed the Memory Hygiene Layer (MHL) as a defense: 100% adversarial detection, 0% false quarantine, 22.6ms p99 latency on CPU hardware — no model retraining required.
Adversarial Memory Injection Attacks on RAG-Based AML Agents
| Finding | Result |
|---|---|
| Baseline AML agent F1 | 0.919 |
| After QTPI attack (50 docs) | 0.014 — 98.4% degradation |
| Log anomalies detected | Zero |
| MHL detection rate | 100% |
| MHL latency | 22.6ms p99 |
The attack exploits the retrieval corpus layer of deployed RAG agents — an attack surface ungoverned by NIST AI RMF, OWASP LLM Top 10, or MITRE ATLAS.
- 🎓 Incoming Cyber Security Analytics student — Mohawk College, Hamilton, Ontario, Canada (September 2026)
- 📄 Paper under peer review at ACM ICAIF 2026, Milan, Italy (notification September 27)
- 🎤 Talk submitted to BSides Toronto 2026, October 3
- 🔬 ORCID: 0009-0008-3483-0025
Adversarial Machine Learning RAG Security and Corpus Integrity LLM Security and Prompt Injection AI Governance in Financial Systems Trustworthy AI for Regulated Environments Financial Crime Compliance Technology
AI Security
Adversarial ML Prompt Injection RAG Security Red Teaming Model Robustness
Machine Learning
Python PyTorch TensorFlow LangChain FAISS Sentence Transformers
Cybersecurity (Mohawk College — September 2026)
Network Security Penetration Testing Digital Forensics Cloud Security SIEM
Data and MLOps
FastAPI Docker AWS PostgreSQL Apache Airflow Power BI
-
AI & ML Engineer — Africa Risk Management and Compliance Partners (ARMC) Production ML systems for Equity Bank Kenya and Commercial Bank of Ethiopia Credit risk agents (88%+ accuracy), AML/fraud detection, graph-based systemic risk modelling
-
Business Intelligence Analyst Intern — Absa Bank Kenya Power BI/Tableau dashboards adopted by 3 departments, 40% faster reporting
-
BSc Economics and Statistics with IT — Maseno University (Second Class Upper Honours, 2024)
-
Data Science Diploma — Moringa School (91.7% average, all Grade A, 2025)
- 🏆 Top 100 Rising AI Developers in Africa 2025 — UNDP, African Development Bank, Microsoft, Meta
- 🎤 Chairperson — Moringa School Alumni Board (13,000+ graduates, 30 countries)
- 🏛️ Governing Council Secretariat — Compliance Society of Kenya
👉 MHL-AML — Full experimental code for the CMPP paper. Run the attack and defense experiments yourself.
📧 ondiekifrank021@gmail.com 🔗 LinkedIn 📄 SSRN Research 🔬 ORCID
If you work in AI security, financial sector cybersecurity, or RAG systems — I would love to connect.