- Interpretability-oriented analysis of model behavior/welfare - Heterogenous Model Swarms in Scalable Oversight / Automated Red Teaming - Evaluation of internal consistency and failure modes - Bio-mimetic memory and state management for long-running LLM agents - Tooling to make agent behavior measurable and auditable