Rich
Harang
I use bad guys to catch math.
Principal Security Architect at NVIDIA.
Working on AI security since it was ML security.
AI Security Architecture
Threat modeling and security design for ML systems at scale
AI Red Team
Thinking about AI/ML as an attack surface
Field Building
Writing, training, and frameworks that help AI security mature as a discipline
From Prompts to PwnsBlack Hat 2025
Exploiting and securing AI agents. My contributions: the AI Kill Chain for modeling attacks, and autonomy levels for scoping agent risk.
Practical LLM SecurityBlack Hat 2024
Grounding LLM security in application security fundamentals. If an attacker can get their data into your LLM, they control the output.
Black Hat Machine Learning
Multi-day adversarial ML training. Evasion, extraction, poisoning—hands-on. Self-paced version available as NVIDIA DLI course.
OpenSSF Model Signing
Helped found the cross-industry working group for signing and validating AI models before handing it off to the signing experts.
CVE Discoveries
Multiple CVEs in AI/ML tooling and infrastructure, as well as other responsible disclosures.
SOREL-20M
20 million malware samples for security ML research. Still the largest public labeled malware corpus.