At a glance
- Agree the definitions first: certified metrics and table descriptions an agent reads before it writes SQL.
- Give each person and each agent a role, with schema, column and row rules, masked personal data and a daily call limit for agents.
- Log every query under a name, so the data team can review what each agent did.
What is data governance for AI?
It covers what the data means, so an agent picks the right table and metric, who may read it down to the column and the row, and a record of what happened, so a person can check an answer or an incident later. Traditional governance answers these in catalogs, policies and access reviews. For AI, the answers also have to be machine-readable and enforced on every call.
How does governance work at query time?
An agent asks for data through a gateway. The gateway knows who the agent is from its token, applies the role’s schema, column and row rules, masks personal data the role may not see, and counts the call against a daily limit. The agent looks up certified definitions before it writes SQL, and the query is validated before it runs. The call, the agent’s name and the tool it ran in go into an audit log.
Why do AI agents need governance of their own?
Agents do not behave like analysts. They run many queries, they run them on schedules, and teams share them in Slack and Notion. Google’s BigQuery MCP documentation recommends a separate identity for agents so access can be controlled and monitored, and Snowflake grants MCP access tool by tool. Those controls work inside each warehouse. A rule set that applies the same way across every warehouse and every AI tool has to live somewhere too, and that is the job of a gateway.
Where do AI governance frameworks fit?
Frameworks such as the NIST AI Risk Management Framework help an organisation decide which risks to manage and who owns them. They do not grant or refuse a query. Use the framework to set the policy, then implement the parts that concern data as roles, rules and logs that the warehouse and the gateway enforce.
How Rig does it
Rig Map holds the definitions: table and column meanings, joins, business terms and certified metrics, drafted from your warehouse and certified by the people who own each number. Gateway enforces access. Each person signs in with their own role, and each agent gets a role, a token and a daily call cap. Roles carry schema access, table and column permissions, PII masking and row filters, and owners can test a role by running SQL as it before granting it. Every call lands in a named audit log, and changes to source systems wait for a person to approve them.
Can Rig help set this up?
Yes. Rig engineers work with your data and IT teams to agree definitions, model roles and connect the first agents, then hand over a setup your team runs. Business owners define each agent’s job and measure its result, and the data team controls definitions and access.
Common questions
What is data governance for AI?
It is the definitions, access rules and audit records that decide what an AI tool may read and let a person check what it did, enforced when each query runs.
How do you make warehouse data ready for agents?
Certify the metrics and table descriptions agents will use, give each agent a role limited to its job, set a call limit, and log every query under the agent’s name.
What does governance look like at query time?
The gateway identifies the caller, applies its schema, column and row rules, masks restricted fields, checks the call limit and logs the query before results come back.
Is row-level access available for agents?
Roles can carry row filters. Tables that filter rows per person need to know which person is asking, so Rig refuses agent queries on those tables.
Sources and product details
Vendor information checked on 20 November 2026. Confirm current editions, licence terms and prices with the vendor before buying.