# Just-in-time access for every AI agent

Source: https://www.hydden.ai/platform/ai-agent-access
Last updated: 2026-11-10

> A Control capability. Hydden adds access for the task, then removes it. The decision is made at runtime for human, machine, and AI accounts.

Control checks each AI agent request at runtime, adds only the access the task needs, and removes it when the task ends.

[Book a demo](https://www.hydden.ai/book-a-demo) Watch the demo

My access

1. Risk check
2. Access granted
3. AI agent executes
4. Access removed

Request access

Reason

Fix failing deploy

Request accessUser's own patterns

Active access

customer-portal-app · write

Ends in 28m 00s

Push to customer-portal-app succeeded

GitHub write on the chris-dev account

Release

My requests

My requests

| Target | Access | Status | Decided |
| --- | --- | --- | --- |
| Oracle | write | Checking |  |
| AWS | read | Verified | Auto-approved |
| Entra | read | Verified | Auto-approved |

- Target Oracle Access write Checking Checking
- Target AWS Access read Verified Auto-approved
- Target Entra Access read Verified Auto-approved

## Eliminate standing access for every AI agent

1. **For years** Identity teams wanted just-in-time access. Most never got there.
2. **Now** AI agents need access decisions in seconds, at a scale people can't approve by hand.
3. **Hydden's approach** Just-in-time access for every AI agent request, decided at runtime and enforced in the systems agents already use.

## How it works

A user asks an AI agent to do a task. Hydden grants only what that task needs, for a set time, then removes it.

1. Justify access With a reason from the user.
2. Approve automatically Based on runtime risk.
3. Grant access For a set time.
4. Revoke and verify When the task ends.

1. **Justify access** With a reason from the user.
2. **Approve automatically** Based on runtime risk.
3. **Grant access** For a set time.
4. **Revoke and verify** When the task ends.

The AI agent's account has no standing access

## How Hydden decides

- **Fast** A System One model decides each request in seconds, using context from the identity system of record to decide with confidence.
- **Contextual** Hydden checks whether the request fits the user's normal working patterns and past behavior.
- **Scored** Every decision carries a confidence score.

[See how decisions work →](https://www.hydden.ai/platform/ai-agent-access/how-decisions-work)

## Why other AI agent access controls fail

Endpoint agents and LLM gateways only watch agents that go through the laptop or gateway they sit on. They leave standing access on the account the agent runs as, where any other route can still use it.

|  | Endpoint agentsSoftware on every host | LLM gatewaysA firewall for AI traffic | HyddenControl where the access lives |
| --- | --- | --- | --- |
| Deployment | Install software on every host. | Route agent through a proxy. | Agentless, and built to work with every AI provider. |
| Maintenance | Maintain software on every host. | Maintain the proxy. | Nothing extra to maintain. |
| Can the agent go around it? | Misses cloud and SaaS. | Easy to go around. | Access is set in each system, so whatever route the agent takes, it only has the rights its account holds. |
| Access after the task | Access stays. | Access stays. | Access is removed. |

## Security teams choose Hydden

- **Gateways are easy to circumvent** A gateway only sees traffic routed through it. Users find ways around them. Hydden sets access in each system, so whatever route the AI agent takes, it can only use the rights its account holds.
- **Zero standing privilege for every account** The same just-in-time control covers human accounts, machine accounts, and AI agents’ own identities. [Learn more](https://www.hydden.ai/platform/control)
- **Decided in real time** Hydden's decisioning engine probes context and returns a yes or no with a confidence score, fast enough to check every request while the agent works. [See how Hydden decides](https://www.hydden.ai/platform/ai-agent-access/how-decisions-work)

## See Hydden control every AI agent

Get your 20-minute walkthrough of the just-in-time AI agent access control

The form could not load. Refresh the page to try again.
