You've built governance on GCP. Now make it work for AI.

For years now, AI budgets have been climbing. C-suites and data teams alike have been chasing the promise of AI. Hires have been made and teams have been formed solely to focus on AI deployment.

And yet, according to Gartner just one in five AI investments show real ROI. It’s easy to point the finger at the models. But the real culprit is often a weak or missing connection between organizations’ existing governance infrastructure and the AI systems now running on top of it.

Many of the world’s leading enterprises have built petabyte-scale data foundations on Google Cloud, with Knowledge Catalog serving as the strategic governance layer across GCP services. Business classifications, quality scans, profiling metrics, and data glossaries are mature, well-maintained, and deeply trusted. As AI agents increasingly query the data this governance covers, connecting that trust context to agentic workflows becomes the next critical step.

As the Enterprise Context Layer, Atlan bridges that by carrying what Knowledge Catalog has already established directly into the AI workflows now running on the same data. Atlan turns business meaning into machine-readable knowledge that AI can understand. Our deepened integration with Google Cloud Knowledge Catalog is where that connection becomes real for GCP customers. Here’s a look at how it works.

Governance is the prerequisite for reliable AI

Google Cloud built Knowledge Catalog to be the governance and management layer for GCP data estates. It handles quality scanning, profiling, data discovery, and the Aspects framework that lets teams attach structured metadata to BigQuery assets at scale. Its foundation is solid and trusted by companies like Walmart, Ford, and Major League Baseball.

But as AI deployments have accelerated, a new need has emerged: a direct path from data stewards’ governance decisions — classifications, certifications, business context — into the systems consuming that data. An AI system without governed, contextualized data will confidently produce bad outputs with no systematic way to catch them.

Atlan is built to deliver governed context at enterprise scale. It’s the universal context layer that captures information about data, semantics, definitions, users, and even the tribal knowledge that lives in people’s heads. And as a Leader in the Gartner® Magic Quadrant™ for Data & Analytics Governance Platforms, Magic Quadrant for Metadata Management Solutions, and the Forrester Wave™: Data Governance Solutions, Atlan has proven its ability to make metadata actionable and trustworthy.

The bidirectional integration between Knowledge Catalog and Atlan connects meaning, governance, and quality. Governance decisions made in Atlan now reach Knowledge Catalog. Quality signals captured by Knowledge Catalog now surface in Atlan. Neither system works in isolation, so users rest assured that data is reliable and governed, no matter how large or complex the environment.

Gartner’s data backs up why this matters: 74% of organizations recognize that data governance tools are key to operationalizing AI governance. Governance leaders hold the infrastructure AI depends on. The question is whether that infrastructure actually reaches AI in a usable form.

What active governance looks like in practice for AI

Most catalog integrations work in one direction: metadata flows in, governance decisions stay put. A data steward certifies a table in the catalog; a developer querying that table in BigQuery sees nothing. The governance work happened, but it didn’t travel.

Atlan’s Knowledge Catalog integration works in both directions.

Inbound: Knowledge Catalog Data Quality and Profiling scan results now surface directly inside Atlan at the column level. Analysts and AI systems see trust signals — quality scores, profiling metrics, custom Aspects — embedded in the asset metadata, not in a separate portal they need to remember to check. The governance work Knowledge Catalog users have already done compounds automatically.

Outbound — Aspects Reverse-Sync: Governance decisions made in Atlan — certifications, PII classifications, business metadata updates — write back to Knowledge Catalog as Aspects, visible in the GCP console to every developer and every downstream system. The catalog stops being where context goes to rest and becomes where context originates.

“Through our integration with Atlan, we’re building the context layer that connects meaning and quality across even the most complex environments. Knowledge Catalog provides context for structured and unstructured data assets on Google Cloud, and Atlan augments it with machine-readable context about definitions, users, data, and semantics from across the data estate. Together, we’re expanding universal context to multi-cloud and hybrid cloud systems, so teams can maximize value from their data and AI.” — Chaitanya Pydimukkala, Product Leader, Google Cloud

This is critical because governance and AI context are the same problem. Gartner has identified semantics, operational state, and provenance as the three components every enterprise needs to get right in order to scale agents reliably. The Atlan-Knowledge Catalog integration delivers all three within the GCP stack: Knowledge Catalog provides the operational state and provenance signals, while Atlan delivers the semantic layer, and bidirectional sync makes it machine-consumable.

In practice, this means when a data steward classifies a set of BigQuery revenue tables as certified and PII-adjacent in Atlan, that classification writes back to Knowledge Catalog as Aspects. It’s visible in the GCP console and present in the asset metadata that Google Agentspace, Vertex pipelines, or any custom LLM workflow will encounter.

This way, safety guardrails are built in: Aspect schema definitions are immutable through Atlan, and deletion is not supported in this release.

What makes this integration different

Teams already familiar with Knowledge Catalog catalog integrations — including those using Collibra’s Knowledge Catalog connector — will recognize the inbound metadata ingestion story. But Atlan’s integration goes significantly further.

Collibra operates as a read-only consumer of Knowledge Catalog metadata. Governance decisions made in Collibra stay in Collibra. A PII classification applied by a steward doesn’t appear in the GCP console; a developer querying that table has no way to see it. Metadata goes in, but doesn’t come out.

Atlan’s integration extends well beyond that baseline. Teams evaluating their options will find a set of capabilities that change the architecture:

Atlan’s initial release targets BigQuery. The underlying architecture is service-agnostic, and GCS, Spanner, and AlloyDB are on the roadmap. Depth before breadth was a deliberate choice — the capabilities that let a data steward do their job take priority over a longer list of passively ingested asset types.

Where to start

There are three priority moves for governance leaders who are evaluating solutions:

First: Audit where your Knowledge Catalog quality scan results currently surface. If analysts and AI systems can’t see them in their working environment — if stewards have to go looking for them — the integration unlocks that value immediately, with no additional governance work required.

Second: Pick one classification that should travel to the GCP layer but doesn’t. PII status, certification, data product ownership — test the Reverse-Sync on one asset class and count how many downstream systems now receive that signal automatically.

Third: Identify one AI use case that’s consuming data without a trust signal attached. That’s the use case most likely producing the unreliable outputs keeping organizations from realizing significant AI value. The integration is how that changes.

Gartner predicts that the organizations that win on enterprise AI by 2027 will be those that invested in governed, semantic context for their data — not simply the ones with access to the best models. The governance infrastructure is already in place for most enterprises on GCP. The question is whether it stays siloed inside a catalog, or whether it becomes active context that AI can actually reach.