---
Title: "What Is Asynchronous Authorization for AI Agents?"
Url: "https://computer.io/glossary/async-authorization"
Topic: "AI safety"
Description: "Asynchronous authorization lets AI agents request human approval for sensitive actions and continue once approved, without blocking. How it works."
---

# What is asynchronous authorization?

Asynchronous authorization is a pattern where an AI agent pauses before a sensitive action, sends an approval request to the right person, and continues only after that person approves, without the user having to be present in the conversation at that moment.

## How it works

1. Agent prepares an action (e.g., refund $500).
2. It requests approval from an authorized person.
3. The approver is notified (email, Slack, app).
4. Agent proceeds or stops based on the response.
5. Everything is logged.

## When to use

Payments and refunds, access to sensitive systems, data deletion and actions above a threshold.

## Benefits

Safety without blocking workflows, and clear accountability.

## FAQ

### What is CIBA?

Client-Initiated Backchannel Authentication, an OpenID standard for approving actions out of band.

### Does async authorization slow agents down?

Only for sensitive actions; others can run immediately.

### Is asynchronous authorization required?

Not always, but it's a best practice for high-risk agent actions.

## Related terms

- [Human in the loop (HITL)](https://computer.io/glossary/human-in-the-loop)
- [AI guardrails](https://computer.io/glossary/ai-guardrails)
- [Role-based access control (RBAC)](https://computer.io/glossary/role-based-access-control)
- [What is async authorization for AI agents, and when should agents work asynchronously?](https://computer.io/answers/when-should-ai-agents-work-asynchronously)

## How Computer helps

- [How computer works](https://computer.io/how-computer-works)

## Further reading

- [Build enterprise ai agents](https://devrev.ai/blog/build-enterprise-ai-agents)