CRAI 2026: The 1st International Workshop on Composable Resilience for Agentic AI

Theme: From Monolithic Brittleness to Composable Trust

As Agentic AI systems are increasingly integrated into critical infrastructures, ensuring their dependability and security has become a pressing concern. Current AI systems often suffer from "monolithic brittleness," where a single failure can lead to unpredictable system-wide consequences. CRAI 2026 aims to address these challenges by advocating for a systems-oriented, compositional approach to resilience.

This workshop focuses on decomposing complex agentic workflows into analyzable modules with explicit interfaces, fault boundaries, and verifiable properties. We bring together researchers and practitioners from AI, systems security, and dependable computing to explore innovative solutions in compositional architectures, fault tolerance, formal verification, and runtime assurance. By bridging the gap between AI dependability and classic DSN principles, CRAI 2026 seeks to establish a foundation for building trustworthy and resilient agentic systems for domains ranging from autonomous driving to industrial automation.

Topics include, but are not limited to:

Compositional Architectures & Software Engineering: modular agentic workflows, verified orchestration layers, sandboxing and isolation, metrics
Resilience, Fault Tolerance & Consensus: fault tolerance, redundancy patterns in reasoning, self-healing, fail safe defaults and graceful degradation
Formal Verification & Runtime Assurance: compositional verification, runtime monitoring of behavioral drift, neurosymbolic verification, model based V&V for agents
Security, Privacy & Ethics: secure agent-to-agent communication, privacy-preserving tool interaction, adversarial defenses, ethical compliance & auditability
Domain-Specific Resilient Agents: Agents in Cyber-Physical Systems: Reliability of agents managing power grids, medical devices, or smart manufacturing units. Dependable real-time decision-making and sensor fusion for multi-agent traffic coordination

CRAI Workshop Co-Chairs:

Eman Hammad, Texas A&M University, USA
Jaewon Kim, Texas A&M Global Cyber Research Institute, USA