# What is an agent approval policy?

Source: https://corespeed.io/glossary/agent-approval-policy

An agent approval policy is a set of plain-English sentences. They say which agent actions run, which ask you first, and which never run. Every write is judged against them and gets one verdict: Allow, Ask or Deny. A default you pick covers what the sentences do not mention. Test policy tries a call before you rely on it.

It rests on the documentation: [Approvals](https://corespeed.io/docs/approvals)

At length, on the blog: https://corespeed.io/blog/what-is-a-plain-english-approval-policy

## What people search for

- ai agent approval workflow
- ai agent guardrails
- agent approval policy
- ai agent approval policy
- plain english approval policy

## Questions

### What happens when the policy says nothing about an action?

The default you pick decides. On Balanced, the default, an ordinary write runs and a destructive one asks you first.

### Can a member loosen the organization's policy?

No. A member's own text can only make their own calls stricter. Where the two disagree, the stricter verdict wins.

### What if the judge cannot answer?

A timeout or an error goes to a separate setting, Ask or Deny. No failure lets a write run.

### Does the judge see your agent's conversation?

No. It sees the policy, the tool, who is calling and, unless you turn it off, the call's arguments.

All terms: https://corespeed.io/glossary (Markdown: https://corespeed.io/glossary.md)

## Related pages

- [Approval gates for AI agent tool actions](https://corespeed.io/solutions/agent-approvals): Set approval policies for writes through CoreSpeed. Review held tool actions while your team keeps using its preferred agents.
- [What is an AI agent spend cap? A budget per key](https://corespeed.io/glossary/ai-agent-spend-cap): An AI agent spend cap is a monthly credit limit on one API key. What happens when an agent reaches it, and how the organization wallet sits above every key.
- [What is human in the loop for AI agents?](https://corespeed.io/glossary/human-in-the-loop): Human in the loop means a person decides named agent actions before they run, under rules written in advance. How the card, the wait and the one run work.

## Sources

- Approvals: https://corespeed.io/docs/approvals
- What is a plain-English approval policy?: https://corespeed.io/blog/what-is-a-plain-english-approval-policy
- Write your first approval policy: https://corespeed.io/blog/write-your-first-approval-policy

Reviewed 2026-11-14.

## Links

- Product overview: https://corespeed.io/ (Markdown: https://corespeed.io/index.md)
- Documentation: https://corespeed.io/docs
- Pricing: https://corespeed.io/pricing (Markdown: https://corespeed.io/pricing.md)
