AI governance

We define the actions an AI system is authorized to take, who is responsible for its use, and how its actions will be reviewed.

Discuss a project

Document the decisions

Written procedures do not always capture how people handle exceptions. In workshops, we ask teams to work through specific scenarios and explain what they check, which sources they use, and when they ask for approval.

We turn those findings into requirements for an AI system.

Define operating requirements

  • Responsibilities. Identify the owner of each workflow and the people authorized to approve changes.
  • Limits. Specify which data and tools the system is authorized to use and which actions need human approval.
  • Exceptions. Define when the system should stop, request more information, or refer a decision to a person.
  • Review. Decide what records to retain and how to investigate an incorrect or unexpected action.

From requirements to a pilot

We deliver a findings register, decision rules, approval requirements, and a pilot plan. Use these requirements to configure and test the pilot.

Read about readiness assessments, risk reviews, and audit trails. Our TN-Proto work explores technical approaches to permissioned event records.