# Action-Time Authorization

> Checking whether a proposed agent action is authorized at the exact moment before the tool call, API request, or side effect executes.

- Category: Evaluation & Safety
- Stage: Emerging
- Also called: Execution-time authorization, Pre-action authorization, Runtime authority control
- Canonical: https://www.joelcsaji.com/ai-dictionary/action-time-authorization

## Plain English

The system asks, is this specific action allowed right now, with these arguments and this source of authority?

## Example

Before a deployment agent runs `terraform destroy`, a runtime guard verifies task scope, target environment, approval state, and policy.

## Why it matters

Prompt-only safety is too weak for agents that act; runtime authorization creates an enforceable boundary before external systems change.

## Related

- [Guardrails](https://www.joelcsaji.com/ai-dictionary/guardrails)
- [Approval Gate](https://www.joelcsaji.com/ai-dictionary/approval-gate)
- [Capability Token](https://www.joelcsaji.com/ai-dictionary/capability-token)

## Sources

- [arXiv: AIRGuard](https://arxiv.org/abs/2605.28914)
- [Capframe: Capability Security](https://capframe.ai/)