Now | Rex Coleman

Last updated: March 16, 2026

Building

Frontier Portfolio v2 — 3 new projects spanning RL, UL, and optimization paradigms applied to frontier security problems. Classical + modern ML architectures (transformers, contrastive learning, RAG).
FP-12: RL Agent Vulnerability — Reward poisoning, policy extraction, and behavioral backdoors on RL-trained agents. 5 algorithms (Q-Learning through PPO + transformer policy). Mapped to 7/10 OWASP Agentic categories. [Starting next]
FP-13: Model Behavioral Fingerprinting — Unsupervised anomaly detection on model activations. 30-combination benchmark (6 detectors x 5 representations). “Antivirus for AI models.”
FP-14: Adversarial Training Optimization — Which optimizer + schedule produces the best robustness-utility tradeoff for LLM safety? Matched compute budget analysis on open-weight models.
govML v2.6 — contract-track profile (36 templates), leakage test generator, A+ quality checklist. 87% adoption across 4 project repos. GitHub

Agent Red-Team Framework — 100% success with reasoning chain hijacking against default-configured agents. 7 attack classes, 19 scenarios.
Adversarial IDS — Feature controllability reduces attack success 35% on network IDS.
Vulnerability Prioritization — ML beats CVSS by +24pp AUC for exploit prediction.
AI Supply Chain Scanner — pickle.load is the SQL injection of ML. 20 findings across 5 projects.
Financial Anomaly Detection — CFA domain features capture 91% of ML signal on synthetic data.
PQC Migration Analyzer — 70% of your crypto isn’t yours to fix.
govML — Open-source ML governance framework.

Georgia Tech OMSCS — Machine Learning specialization (4/10 courses complete)
CS 7641 ML complete: top-1% rigor across SL, OL, UL, RL — those patterns now govern all frontier projects