Orchestrating Resilient and Compliant Agents

Challenge

Agentic AI systems are rapidly entering industrial software environments, where autonomous agents, often from organisations, interact and make decisions in a systems-of-systems context. While this enables powerful new applications, it also introduces major challenges in safety, security, scalability and regulatory compliance. Companies struggle to verify and control the behaviour of interacting agents, especially in the context of the EU AI Act, which requires AI systems to be safe, transparent and traceable. Current tools and standards do not provide a systematic way to specify, verify and demonstrate correct agent behaviour across complex multi-agent systems.

Project goal

Through this Strategic Basic Research project (ORCA_SBO), we aim to enable resilient, secure and AI Act–compliant multi-agent systems through a contract-based approach. The project develops domain-specific languages and automated tooling that allow companies to formally specify what “safe, secure and correct” behaviour means for agentic AI systems, and to verify this behaviour before and during deployment.

By reducing reliance on scarce experts and automating large parts of agent verification, we help companies onboard agentic AI faster, scale more confidently, and reduce the cost and risk of compliance.