AI agent context layers: what happens when ontology fragments

TL;DR: Governed context improved agent accuracy from 62% to 92% in a KU Leuven test, while ungoverned agents answered confidently wrong on definition-dependent questions and burned more compute, according to Collibra. The deeper issue is assumption collapse: access control and data governance models built for fixed truth sources fail when agents reason across fragmented ontologies at runtime.

NHIMG editorial — based on content published by Collibra: Data has gravity. Context does too

By the numbers:

• Without it, 62%.
• Only 13% of organisations feel extremely prepared for the reality of agentic AI despite the majority racing toward autonomous adoption.
• 70% of organisations grant AI systems more access than they would give a human employee performing the exact same job.

Questions worth separating out

Q: How should security teams govern AI agents that reason across multiple data platforms?

A: Security teams should govern the meaning layer, not just the access layer.

Q: Why do fragmented ontologies create risk for AI agents?

A: Fragmented ontologies give agents multiple, incompatible definitions of the same business term.

Q: What breaks when AI agents rely on platform-specific catalogs for context?

A: What breaks is enterprise consistency.

Practitioner guidance

• Map high-value business terms across platforms Identify where customer, revenue, active user, and similar terms diverge across CRM, warehouse, SaaS, and legacy systems.
• Test agent output with and without governed context Run the same questions through your preferred model with governed definitions, lineage, and quality signals enabled and disabled.
• Separate platform metadata from enterprise meaning Do not assume a cloud catalog automatically gives you a neutral enterprise ontology.

What's in the full article

Collibra's full blog post covers the operational detail this post intentionally leaves for the source:

• The KU Leuven test design, including how the anonymised data was structured and how the Claude agent was queried.
• The specific examples of incorrect reasoning, including the active tester case and the definition retrieval path that corrected it.
• How Collibra positions ontology, lineage, and quality signals inside its enterprise AI control plane.
• The practical argument for a neutral context layer across Databricks, Snowflake, BigQuery, and on-prem systems.

👉 Read Collibra's analysis of context gravity and AI agent governance →

AI agent context layers: what happens when ontology fragments?

Explore further

View Full Forum →  |  NHI Foundation Course →