Blog

AI Governance with Atlan: AI Use Cases, Risk Assessments, Workflows & Shadow AI Governance

Atlan home screen listing recently verified AI Risk Assessment terms, including Digital Twins for Clinical Trials

The Tavro team developed an AI Governance prototype in Atlan. We reused the existing operating model with assets and added custom attributes and relations.

AI Use Cases

As discussed in an earlier blog, a digital twin may be a virtual replica of a particular patient that reflects the unique genetic makeup of the patient or a simulated three-dimensional model that exhibits the characteristics of a patient’s heart. Digital twins may be utilized to accelerate clinical trials and reduce costs in the life sciences industry. The Tavro team implemented an overview of the Digital Twins for Clinical Trials AI Use Case in Atlan.

Atlan glossary page with the AI Use Case Overview for Digital Twins for Clinical Trials, showing description and business problem

AI Risk Assessments

We conducted an AI Risk Assessment for the use case with Atlan. Digital twins have the potential to introduce bias risks based on the algorithms and the underlying data sets. We documented the bias risk assessment and a mapping to the associated regulations in Atlan.

Atlan bias risk rating set to Medium, with rationale on biased algorithms and training data and links to EU AI Act Article 10 and NYC Local Law 144

We also documented the privacy risks in Atlan.

Atlan privacy risk rating set to High for the digital twins use case, citing sensitive health data and GDPR, plus a High security risk rating

We documented other dimensions of AI risk including Reliability, Accountability, Explainability and Security in Atlan. For the sake of brevity, I have not included those screenshots here.

This use case would likely be classified as High Risk based on the Medical Device category of Article 6 of the EU AI Act.

Overall risk rating of High for the AI use case, classified as a high-risk system under Article 6 of the EU AI Act for medical devices

AI Risk Assessment Workflows

We configured an AI Risk Assessment workflow in Atlan to route the AI Risk Assessment to the appropriate parties for approval.

Atlan governance workflows screen showing the AI Risk Assessment Approval Flow and its rule-based approval steps

The screenshot below shows the AI Risk Assessment in Approved status based on approvals from the Operational Risk Management Committee (ORMC) and the AI Governance Council.

Approved AI Risk Assessment request in Atlan, signed off by the ORMC and the AI Governance Council

Shadow AI Governance to Ingest Metadata from ServiceNow CMDB and Tavro Agents on Hugging Face to Highlight COTS Apps with Embedded AI

In an earlier blog, I discussed Shadow AI Governance and the Tavro agents. As part of the current exercise, we ingested metadata around the Commercial-off-the-Shelf (COTS) apps into Atlan. This information includes metadata such as Application Name, Privacy Policy URL, Data Specifically Excluded from AI Training, Embedded AI and Opt-Out Option.

The screenshot below shows Atlan before running the integration with the Tavro agents. The catalog only contains one AI Use Case (Digital Twins for Clinical trials) and one application (Google Product Services).

Atlan glossary before the integration, containing only Digital Twins for Clinical Trials and Google Product Services

After running the integration with Atlan API, Atlan contains a broader list of applications including Actimize Xceed including metadata in the right panel.

Atlan glossary after the integration, listing more applications and Actimize Xceed metadata showing Embedded AI set to Yes

Conditional Logic with Atlan API to Auto-Create AI Use Case and AI Risk Assessment Objects

We implemented conditional logic in the Atlan API to auto-create AI use cases only for applications with embedded AI. In this case, we created an AI use case object in Atlan for Actimize Xceed because Embedded AI = “Yes.”

Atlan auto-creates an AI Use Case under Actimize Xceed because its Embedded AI attribute is Yes

We also implemented conditional logic in the Atlan API to auto-create AI Risk Assessment objects where Data Specifically Excluded for AI Training = “No.” Obviously, this logic is configurable.

Actimize Xceed AI Use Case in Atlan with Data Specifically Excluded From AI Training set to No, triggering a risk assessment

This is a basic AI Governance configuration in Atlan with more to come!

NewerCleansing Software Inventory Data in ServiceNow CMDB to Comply with FFIEC Guidelines at a Regional Bank
← All posts