Kubrick’s Collibra Centre of Excellence has analyzed the features released in 2026, and how to implement them correctly to realize their value.
The research is informed by our delivery experience, which was recognized by Collibra who named Kubrick Regional Systems Integrator, EMEA, of the Year, 2026.
The value of a new feature depends on the problem it helps you solve. We examine where the technology is useful, what implementation requires and which prerequisites deserve attention before you commit.
Explore what Collibra latest developments mean for your Data and AI Governance program
- 01
Control Tower
Metadata standards that you can measure.
- 02
Outbound metadata
Governed context where your teams work.
- 03
Data Marketplace
A catalog people can confidently use.
- 04
AI Command Center
AI oversight that supports better decisions.
- 05
Collibra Model Context Protocol (MCP)
Business context for your AI workflows.
Control Tower
Metadata standards that you can measure
A published standard has limited value if nobody can show where it is met. Control Tower gives governance teams a repeatable way to identify gaps and prioritize action.
Why it matters
As a Collibra catalog grows, checking metadata standards by hand becomes a recurring drain on stewardship capacity. Exports provide a snapshot, but teams still need to interpret the results and find someone to act. Control Tower runs scheduled checks against metadata in Collibra’s knowledge graph. Scheduled checks reduce the effort of repeatedly establishing which assets meet the agreed standards. Governance teams can spend more time resolving failures and report whether adherence is improving over time.
2026 development: Collibra Control Tower Dashboard in action
Where the feature is strongest
Controls can follow relationships across the metamodel, with hourly, daily or weekly execution. Configurable email alerts route new failures to relevant resource roles.
Shared dashboards show control status and trends, with category and severity filters to support prioritization. The dashboard gives governance teams a consistent view of control results, helping them track whether fewer assets are failing the agreed checks.
Client experience: a prioritized metadata backlog
Kubrick implemented minimum metadata checks across a Global Insurer’s business semantic assets, including Data Concepts, Business Terms, Synonyms, Acronyms and Data Products.
The team agreed control categories during configuration, so the first results formed a prioritized remediation backlog. The governance team could identify which assets failed, the agreed standards and use subsequent runs to track changes in the backlog.
Kubrick’s tips
Automated checks need human judgment. A populated field may pass a completeness check while containing an incorrect or unhelpful value, so use sampled reviews to assess what automation cannot.
Control Tower is part of the core platform, but Collibra states that executions consume Collibra Units from 1 October 2026. Choose the scope and frequency of checks with both governance needs and cost in mind.
Controls are most useful when stewards can see why an asset has failed and what needs to change. Before scheduling a control:
Outbound metadata
Governed context where your teams work
Metadata creates more value when it reaches the people making decisions about data. Outbound synchronization takes selected Collibra context into the platforms they already use.
Why it matters
An engineer working in a data platform may never open the catalog. If governed context stays inside Collibra, teams can duplicate documentation or make decisions without seeing the standards already agreed. Outbound metadata synchronization lets selected Collibra attribute values populate tags on corresponding platform objects. Teams can maintain approved metadata in Collibra and publish selected values to the corresponding Databricks or Snowflake objects. This reduces duplicate maintenance and gives users access to that context while they work with the data.
Databricks and Snowflake outbound integrations reached general availability in September.
Description mapping remains in preview.
Where the feature is strongest
Mappings allow teams to select the metadata they want to distribute. Both integrations process pending outbound events on a five-minute cycle. That creates a repeatable route for sharing approved context without maintaining each value manually in multiple places.
The practical opportunity is consistency: the business definition or usage guidance governed in Collibra can be available closer to the data it describes.
Client experience: context for Databricks Genie
For a global insurer, Kubrick used outbound metadata synchronization to publish governed definitions, criticality and security classifications from Collibra to Unity Catalog tags. We also provided a framework for configuring Genie to use this metadata when interpreting business questions. This made context previously absent from Unity Catalog available to Genie and helped Genie interpret organization-specific synonyms and acronyms using their governed meanings, reducing its reliance on assumptions.
Kubrick’s tips
Check the prerequisites for your target platform before configuring outbound synchronization:
- Databricks: assets must have been ingested through the Unity Catalog integration
- Snowflake: asset matching relies on JDBC-ingested metadata, and the target tags must already exist
- Think through the flow of metadata, how to handle conflicts, changes, and what you want to propagate – sometimes too much context is hindering depending upon the challenge.
Description mapping also has exclusions, including columns of database views. Confirm that the metadata you intend to publish is supported, then test how synchronization handles mapping changes, deletions and conflicts with local edits. Assign an owner to monitor failures and resolve discrepancies.
Be clear about which platform owns each value: an edit in the target platform should not be assumed to update the same field in Collibra. Likewise, publishing tags makes context available to AI assistants, but using it effectively requires configuration and testing. For Genie, our setup framework helps teams put that context to use.
Data Marketplace
A catalog people can confidently use
A well-populated catalog does not automatically help someone choose the right data. Data Marketplace makes the consumer journey an explicit part of governance.
Why it matters
Collibra’s revamped Data Marketplace experience allows users to quickly discover, understand, and access trusted and published assets – Data Products, Data Concepts, Data Sets, Reference Data – whichever your users would go to your marketplace for!
If they’re going there to search and find a product, the experience is now tailored to allow a simpler and more friendly UX aligned to categories/domains. If they can’t find what they want, the marketplace now has custom resource tiles allowing workflow triggers – “Suggest a Data Product”!
2026 development: the redesigned Data Marketplace became generally available in release 2026.08.
Where the feature is strongest
Collibra’s new enhancements to Data Marketplace allow a far greater level of customization and tailoring to your user’s requirements. Collibra have taken on the feedback regarding the degree of customization required from visuals you can upload, assets you can set as default (and assign to categories/domains!), and workflows you can trigger from the marketplace with custom resource cards.
A greatly enhanced portal you can direct your users towards and tailor their around.
Client experience: designing for consumption
For a large European business, Kubrick organized the Marketplace around the questions consumers needed to answer when choosing data. This informed which assets to publish, how to group them and what information to show on each card. The design also aligned discovery with the organization’s access process.
In a separate data-product program, Kubrick defined how products progressed through ownership and approval into consumption. Marketplace provided the consumer-facing layer, presenting approved products with consistent business context and usage information.
Kubrick’s tips
A better interface makes metadata, data products, contracts etc. easier to find but also makes gaps more visible. Give someone responsibility for keeping published content useful, with clear publication criteria covering meaningful descriptions, accountable owners and current usage information. Utilizing other features such as Control Tower alongside your marketplace content will allow to you ensure a governed and quality data marketplace.
Before broadening the Marketplace’s scope:
AI Command Center
AI oversight that supports better decisions
As AI agents take actions across business systems, leaders need a clearer account of what is in use and who is responsible. A governed inventory is the starting point for that oversight.
Why it matters
Periodic reviews become harder to sustain as teams deploy more AI systems. A register can quickly drift away from the technology in use, leaving owners and risk teams with different views of the same estate. AI Command Center brings use cases, models and agents into a shared governance view. Supported integrations and relationships connect technical assets with business context. When an agent comes forward for approval, reviewers need to understand its purpose, who owns it and what evidence supports deployment. Connecting those records helps them identify what is missing and decide whether further review is needed.
2026 development: AI Command Center brings an expanded registry, dashboards and AI Trust Scores to Collibra’s AI governance experience.
Where the feature is strongest
A common registry and dashboards give technical teams, governance functions and leaders a shared starting point. Assessment templates aligned to frameworks such as the EU AI Act and NIST AI RMF can help structure evidence collection. AI Trust Scores can help identify assets that need attention. Reviewers still need to examine the contributing factors to understand what the score says about that asset and which gaps require action.
Kubrick’s tips
A high Trust Score indicates the extent to which the recorded evidence meets your configured governance criteria. It does not establish that an agent produces accurate answers or behaves reliably. Deciding whether to trust an agent for a particular purpose still requires an evaluation of its behavior and performance.
Before rollout:
Client experience
Kubrick have implemented AI Command Centre in their own organisational estate connected across Databricks to manage AI use case capture, AI Models, and impact on table training.
Utilizing this has allowed us to govern our own AI estate and ensure governance is transparent and useful.
Collibra MCP
Business context for your AI workflows
AI can generate a confident answer while misunderstanding the data behind it. Collibra’s Model Context Protocol (MCP) offering gives compatible tools a route to governed business context.
Why it matters
Business definitions, relationships and lineage can change how a dataset should be interpreted. When an AI workflow cannot access that context, it may make assumptions that are difficult for a user to spot. Collibra’s MCP tools let compatible clients retrieve metadata and, where authorized, perform supported actions. This creates a route to use governed context inside an AI workflow. The value still depends on the quality of the catalog and on how the client selects and uses the available tools.
Where the feature is strongest
Read tools can retrieve asset details, explore lineage and connect business terms with physical data. Documented integrations include Claude, ChatGPT, Databricks AI Playground and Agents, and Snowflake Cortex. The hosted server uses OAuth and respects the connecting identity’s permissions. Collibra also offers a local, community-supported server. These are different deployment choices with different setup and support arrangements.
OAuth 2.0 authorization-code support arrived in 2026.06, extending an MCP offering already documented in 2025.11.
Client experience: semantics and discovery
At a large investment bank, Kubrick has helped use Collibra MCP to match business terms and columns using asset metadata. This provides a basis for semantic mapping, with proposed matches subject to validation. At a different large central bank, Collibra MCP supports metadata discovery for users and agents. Bringing definitions and relationships into that workflow can reduce the need to navigate the catalog for routine questions.
Kubrick’s tips
Grounding gives the model evidence to work with, but it can still misinterpret that evidence. Require source references so users can check the basis of an answer, and test representative questions with subject-matter experts before relying on the results.
When planning implementation:
Conclusion
Collibra have truly listened and incorporated features which users have been asking for – every one of these developments is worth a look.
However, none of them does much on its own.
- Control Tower is only as good as the standards behind it.
- Outbound metadata needs high quality metadata to push.
- Marketplace needs publication criteria and an access route that works.
- AI Command Center needs named reviewers and approvers aligned with an AI Governance approach.
- MCP needs a catalogue you would be comfortable having an AI quote back to you.
The feature is the easy part. The ownership, testing and running costs around it decide whether it pays off.
Start with the problem, rather than the release notes. Pick the gap that is causing you the most pain today. Perhaps metadata standards nobody can evidence, AI assistants guessing at your business terms, or an AI estate nobody can fully account for. Then try the relevant feature against it, with real assets, real users and an agreed idea of what good looks like.