Giskard Hub and Collibra AI Governance sit on either side of the same problem: Giskard generates the evidence that an AI system actually behaves safely and reliably (groundedness, hallucination rates, robustness, red-team results across probes), while Collibra is where that evidence needs to live for governance, risk, and compliance teams to act on it. Without an integration, that evidence tends to stay siloed in engineering tooling — visible to the team that ran the scan, invisible to the risk officers, auditors, and business stakeholders who are accountable for the AI system under frameworks like the EU AI Act. The Giskard-to-Collibra skill closes that gap directly: it takes the latest Hub scan for a given agent, automatically creates or updates the corresponding AI Agent asset hierarchy in Collibra, and pushes per-probe pass/fail metrics. Because re-runs are idempotent, this can run on every scan without creating duplicate assets, effectively turning Collibra into a living, continuously updated system of record for AI quality and safety posture.
For joint customers, the practical benefit is that AI governance stops being a manual, point-in-time exercise and becomes something closer to continuous compliance. Risk and compliance teams get an up-to-date view of every AI agent’s test coverage and failure modes directly inside the catalog they already use for data and model governance, without needing to chase engineering for the scan results. Engineering teams, in turn, don’t have to maintain a second reporting pipeline — the same red-team and RAG evaluation suites they run for their own quality assurance become the audit trail that feeds Collibra. This is especially valuable as organizations scale from one or two AI agents to dozens: the integration gives a consistent, automated way to keep governance metadata in sync with reality, which is exactly the kind of traceability regulators and internal risk committees are starting to require.
