Permission-scoped access
Connect only the accounts and permissions an agent needs for its specific job.
Security & Control
AI agents should have enough access to work — not unlimited access.
Security for agentic workflows starts with clear scope, permission boundaries, human approvals and visibility into actions across connected systems.
AI agent security
human in the loop
agent permissions
Security principles
Guardrails for agents that can take action.
Human approvals
Place review checkpoints before sensitive or high-impact actions.
Credential handling
Keep sensitive connection tokens server-side instead of exposing them in the customer interface.
Activity visibility
Design workflows so teams can see what agents did, where they acted and when a human took over.
Clear agent boundaries
Define what an agent may do, what it must never do and when it must escalate.
Least-privilege mindset
Start narrow and expand access only when the workflow proves it requires more.
Human control
Autonomy should increase only where the workflow earns it.
Start with low-risk, high-value tasks. Use approval checkpoints for actions such as sending sensitive messages, changing important records or committing financial decisions. Increase agent permissions only after the workflow is stable and observable.
Deployment checklist
✓Define the agent’s allowed job
✓Limit connected permissions
✓Set escalation rules
✓Review activity and exceptions
Plan agent workflows with clear boundaries.
Choose a business job, define its permissions and keep people in control of exceptions.