Collibra brings runtime governance to enterprise AI agents

Runtime governance is gaining importance as autonomous AI agents shift from merely answering questions to executing actions within enterprise systems. To manage both the business context and boundaries of these agents, Collibra BV has introduced Live Map, Maestro, Guardian Agents, and Agent Contracts for runtime governance. CEO and co-founder Felix Van de Maele emphasizes that the costs of manual verification, rework, and risk—referred to as the "hallucination tax"—have significantly increased.

The shift from human judgement to automated governance is a result of the absence of human oversight in AI-driven processes. Van de Maele highlighted that shared knowledge layers, including a variety of context sources, enable agents to understand their environment, explain their responses, and reuse knowledge across tasks. While graphs are a crucial context layer, Van de Maele does not consider them a one-size-fits-all solution, suggesting that different retrieval mechanisms may be more suitable for specific tasks.

Live Map preps reusable context from unstructured documents, ensuring quick access to information without the need for the model to parse data repeatedly. Collibra's Knowledge Graph, built on Neo4j, facilitates the traversal of relationships among enterprise assets. Maestro automates governance tasks previously performed manually, enabling customers to manage AI workloads more effectively.

By defining permitted behavior through Agent Contracts and deploying them in AI gateways, Guardian Agents ensure that agents adhere to established guidelines before taking any action.