Trustworthy Autonomous & Agentic AI

AI systems are increasingly composed of autonomous agents that retrieve external knowledge, call software tools, coordinate with other agents, and take actions with real-world consequences. Trust in these systems requires more than predictive accuracy — it requires explainability, accountability, and verifiable alignment between agent behavior and human intent.

This thrust focuses on:

• Explainability and interpretability methods for autonomous and agentic AI decision-making
• Accountability, auditability, and provenance tracking across multi-agent and tool-using pipelines
• Safety and alignment evaluation for retrieval-augmented and tool-invoking large language model systems
• Resilience of autonomous agents operating in uncertain, adversarial, or partially observable environments

The aim is to develop principled foundations and practical tools for deploying autonomous AI agents that remain transparent, controllable, and trustworthy as their autonomy increases.