AI Safety & Governance Lead

RLHF, DPO, and constitutional AI approaches to value alignment.

Adversarial prompting, automated red-team suites, and vulnerability disclosure.

Documenting capabilities, limitations, intended use, and known failure modes.

Disparate impact testing across demographics and protected attributes.

NIST AI RMF, EU AI Act, and ISO/IEC 42001 compliance mapping.

Harm severity scales, likelihood scoring, and risk register maintenance.

Attention probing, activation patching, and circuit analysis.

Detection, containment, root-cause analysis, and postmortem for AI failures.

Training data tracking, consent management, and deletion workflows.

Input sanitization, instruction hierarchy, and indirect injection mitigation.

Drift detection, output distribution tracking, and alerting thresholds.

Differential privacy, federated learning, and synthetic data generation.

Audit trails, decision logs, and accountability matrices for AI systems.

Translating technical risk findings for executives, regulators, and users.