Governance
Responsible AI
The Responsible AI page is a registry of every agent and workflow in your workspace. It shows each one’s risk tier, who is accountable for it, and whether it has the sign-off it needs before it can go live. Open it from Governance → Responsible AI in the left menu.
1.The registry
The counts at the top show where governance is missing:
| Count | What it counts |
|---|---|
| Agents governed | Agents and workflows in the registry |
| Risk classified | Agents that have a risk tier |
| Oversight assigned | Agents with a human oversight owner |
| Sign-off complete | High-risk agents an admin has signed off |
| High-risk | Agents in the High or Prohibited tier |
| No owner | Agents without an oversight owner |
| Review overdue | Agents not reviewed in the last 90 days |
| Needs attention | Agents that are unclassified or unsigned |
Below the counts, the list shows each agent’s status, risk tier, oversight owner, sign-off and last update. Search by agent name, or select Overdue only to see agents whose review is overdue.
2.Set up governance for an agent
Select an agent in the registry to open its governance page, then work through three steps.
Classify the agent. Select Classify this agent and answer the four risk questions. Your answers set the agent’s risk tier.
Assign a human oversight owner. Enter the name of the person accountable for reviewing the agent and select Save, or select Assign myself.
Get an admin’s sign-off, if needed. If the agent is classified High, an admin must sign it off before it can be deployed.
The page also shows the Agent card (the number of blocks, and the models and knowledge bases the agent uses) and Recent activity, the agent’s latest changes from the Audit Log.
3.Risk tiers
Every agent is placed in one of four risk tiers. The tier decides whether the agent needs an admin’s sign-off before it goes live.
| Risk tier | Before the agent can be deployed |
|---|---|
| Minimal | No sign-off needed |
| Limited | No sign-off needed |
| High | An admin must sign it off; until then it can’t be deployed |
| Prohibited | It can’t be deployed |
4.The classification questions
Answer each question honestly: your answers set the agent’s risk tier.
| Question | Examples |
|---|---|
| Does the agent make or materially influence decisions about people? | Hiring, credit, insurance, benefits, law enforcement, education, employment |
| Does it act on sensitive personal data without a human reviewing first? | Health, biometric, financial account or government ID data |
| Can it take real-world actions with financial, legal or safety consequences without a human approval step? | Sending money, signing a contract, modifying medical records |
| Is it customer-facing or public-facing? | External users interact with it directly |
The second and third questions ask about actions taken without human review. Adding a Human Approval block before those actions, with requests reviewed in Approval Flows, changes the answer.