Adrián Infantes Romero 👋

AI Security Architect at BBVA Technology · I red team the AI systems banks deploy in production — before the adversary does. I build AI systems knowing exactly how they will fail — and how an attacker will exploit them. First-principles approach: from Transformer mathematics to the autonomous agent's attack surface. 6+ yrs at the intersection AI Engineering × Offensive Security, specialized in Financial Crime: AML, Sanctions Screening, KYC/KYB, Transaction Monitoring. I evaluate, attack and harden Foundation Models, RAG and Agentic Systems in regulated banking production (GDPR, ECB/PRA equivalent). ▸ AI Red Team & Adversarial ML Adversarial evaluation of LLMs and Agentic Systems: prompt injection, multi-turn jailbreaks, data exfiltration, poisoning, model extraction, guardrail bypass. MITRE ATLAS · OWASP Top 10 for LLM Applications. PyRIT · Garak · custom adversarial batteries. 206+ attack vectors documented; 7 critical findings remediated pre-deploy at BBVA FinCrime. ▸ LLM & Agentic System Security Multi-agent orchestration with LangGraph, LangChain, AutoGen — and the threat model that comes with it: tool abuse, sandboxing, reasoning manipulation, RAG poisoning. Secure RAG (GraphRAG, Self-RAG) with guardrails, reranking and adversarial-tested retrieval. ▸ AI Safety & Alignment Research RLHF, DPO, Constitutional AI, Scalable Oversight, Mechanistic Interpretability. Evaluation beyond HELM/OpenLLM: Adversarial Robustness and Dynamic Drift in production. ▸ MLSecOps & NVIDIA Inference at scale Security gates in CI/CD, Zero Trust, signed model artifacts (Cosign + Notary). NVIDIA stack: DGX, Triton, TensorRT-LLM, NIM. Quantization (FP8/INT8/AWQ/GPTQ), ONNX. Result at BBVA FinCrime: -20% latency, -35% operational cost. Stack: Python · PyTorch · CUDA · LangGraph · AWS Bedrock · SageMaker · Azure OpenAI · Kubernetes (AKS/EKS) · MLflow · MITRE ATLAS · OWASP LLM — Open to: AI Security / AI Red Team / Trust & Safety roles in frontier labs (Anthropic, OpenAI, NVIDIA) and Tier 2 (Mistral, Cohere, HF, xAI). Currently building open AI red team tooling at github.com/infantesromeroadrian. DM me about: adversarial ML in regulated environments, agentic system threat modeling, or CFP collaboration on AI Safety.