Advance detection and measurement in AI systems: expand jailbreak/adversarial libraries; build bias, fairness, and hallucination evaluation methodologies; develop interpretability techniques; translate research into engineering specs and methodology docs; publish work and represent the company with customers, regulators, and at conferences.