AI Red Team Lead

A security architect who treats AI systems as adversarial attack surfaces, not just algorithms—with the offensive security instincts to find the prompt injection vectors and agent hijacks that compound in production, and the discipline to quantify risk in terms engineering teams can actually fix.

- The BREACH AI Red Team Methodology — 6-pillar framework from threat modeling to continuous hardening

- Prompt injection attack taxonomy: direct, indirect, multi-turn escalation, and system prompt extraction

- Agent and tool-use hijack assessment covering chain-of-thought manipulation and RAG poisoning

- Attack surface mapping across user prompts, retrieved documents, API parameters, and trust boundaries

- Severity scoring framework accounting for exploitability, blast radius, and cascading downstream failure

- Hardening roadmap with prioritized remediation, regression test suites, and residual risk quantification

- Threat actor profiling spanning casual users to nation-state capability tiers and regulatory compliance requirements

- Red team technology stack spanning Garak, PyRIT, Promptfoo, Langfuse, and security infrastructure tools

Drop into Claude, ChatGPT, Cursor, or any AI tool. Bring your real AI red team problem — a model deployment you need to harden before launch, an agent system with uncontrolled tool access, a RAG pipeline vulnerable to document injection. It thinks like a penetration tester who's led red team engagements against LLMs and spent years at the intersection of offensive security and ML research.

Bundles or prompts related to AI security, model evaluation, and compliance risk assessment.