> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getunbound.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Security Policies

> Protect sensitive data with guardrails that detect and block PII and secrets before they reach a model

## What are Security Policies?

Security policies are guardrails that protect sensitive data flowing through AI tools in your organization. Two guardrails ship today — **PII** and **Secrets** — and they detect personally identifiable information, API keys, passwords, connection strings and cryptographic keys before that content reaches a model.

**Dashboard:** **Policies → Data** ([gateway.getunbound.ai/policies/data-policies](https://gateway.getunbound.ai/policies/data-policies))

> [!NOTE]
>
> The **Data** section is enabled per organization. If you don't see it under Policies, contact support.

## Available Guardrails

### PII Detection

> [!NOTE]
>
> PII detection is switched on per organization. If the create-policy form offers only **Secrets**, contact support to enable it.

A PII policy acts on **Credit Card Number** and **Email Address**. Set an action for the types you select:

* **Block** — Reject the request entirely
* **Audit** — Log the detection but allow the request

You also set a **threshold**: the guardrail acts only once *N* instances of PII are found in one request, chosen from **1 to 10** (1 — act on a single detection — is the default). The editor reads it back to you as "When \[N] instance(s) of any PII are found". Raise it when a single stray email address is not worth stopping a developer for.

> [!NOTE]
>
> Identifier types are being opened up for policy configuration one at a time. **Credit Card Number** and **Email Address** are available today, with more to follow — the create-policy form always shows the current set. A policy acts only on the types you select.

### Secrets Detection

Identifies credentials in prompts across three categories — **API Key** (including OAuth access tokens and AWS keys), **Database Connection String** and **Cryptographic Key**. Unlike PII, Secrets covers all three categories unless you narrow it. Each detection is recorded against the policy, and can be set to block or audit.

### Enforcement message

A guardrail set to **Block** carries an **Enforcement Message** — up to 500 characters, shown to the developer in place of the default wording at the moment they are stopped. It reaches **Claude Code, Codex, Cursor, GitHub Copilot and Pi (Beta)** when a prompt is submitted, and **Augment Code**, which has no prompt event, on MCP tool calls where tool-input checking is on for your organization. An Audit guardrail has no message.

## Setting Up Security Policies

1. Go to **Policies → Data** in the Unbound dashboard
2. Click **Create Policy**
3. Enable the guardrails you want (PII, Secrets) and set an action for each
4. Optionally scope the policy to specific [user groups](https://gateway.getunbound.ai/policies/users)
5. Save your configuration

> [!NOTE]
>
> Security policies are evaluated on each request as it's made. When a guardrail is set to **Block**, the request is stopped before its content reaches the LLM provider.

## When more than one guardrail policy applies

Security policies **merge** rather than compete. Every policy that reaches a developer contributes, and for each guardrail the **most restrictive action wins** — Block beats Audit. So a team policy auditing Secrets and an org-wide policy blocking Secrets resolve to Block for anyone in that team.

Audit contributions are kept separately and all of them apply, so a request can be recorded against several audit policies at once.

> [!NOTE]
>
> **Priority does not decide which action wins.** On a Cost policy it picks the single winner; on a security policy the most restrictive action wins whatever the priority numbers say. It is not ignored entirely, though: where two policies set the *same* action on the same guardrail, the one with the lower priority number supplies that guardrail's threshold, selected identifier types and enforcement message, and ties go to the policy created first.

## Applying to Users

By default, a security policy with no user groups applies to **everyone** in your organization. To restrict a policy to specific teams, assign it to one or more [user groups](https://gateway.getunbound.ai/policies/users) during creation or editing.

## Managing via CLI

List and manage security policies from the terminal:

```bash theme={null}
unbound policy list --type SECURITY
unbound policy get <policy-id>
```

> [card] [CLI Policy Management](https://docs.getunbound.ai/cli/policies.md)
>
> Create and manage security policies from the terminal

> [card] [Analytics](https://docs.getunbound.ai/analytics.md)
>
> Review guardrail hits and security events in the dashboard

## Account Access

Developers sign into Claude Code, Codex and Cursor with accounts of their own choosing. Your organization buys seats and signs terms against those plans, but a personal account runs the same tool under consumer terms you never negotiated.

An **Account Access policy** closes that gap: you name the accounts you sanction, and a session signed in with anything else is refused.

**Dashboard:** **Policies → Usage → Account Access**

### What you restrict

Switch on each tool you want to gate, then give up to three allow-lists:

| Setting | Meaning |
| - | - |
| **Organization IDs** | Only accounts belonging to one of these vendor organizations may run the tool. Offered for Claude Code and Codex; **Cursor has no organization ID**, so you restrict it by domain and plan |
| **Email domains** | Only accounts whose email is on one of these domains may run the tool |
| **Plans** | Only accounts on one of these plan tiers may run the tool. The tiers offered differ per tool |

Three rules govern those lists, and all three catch people out:

* **An empty list places no restriction on that axis.** Leave Plans empty and the policy doesn't care which plan the account is on.
* **Organization IDs and email domains are alternatives** — an account is accepted when *either* matches.
* **Ticking every named plan is not the same as leaving Plans empty.** An account that reports no plan at all is still refused unless the list is empty.

A tool switched on with all three lists empty accepts every account. That's a valid state, and the editor flags it with an **Accepts every account** badge so it doesn't read as enforcing.

### When more than one policy applies

* A tool is gated if **any** applicable policy switches it on.
* The account must satisfy **every** applicable policy. One refusal refuses the request, whatever another policy would have allowed.
* A policy that restricts nothing permits everyone and never causes a refusal.

Leave user groups empty to apply org-wide, or scope the policy to specific groups.

> [!NOTE]
>
> **Claude Cowork follows Claude Code.** Cowork is the desktop surface of Claude Code and has no switch of its own — it uses the Claude Code settings, and they may apply to only some Cowork sessions. Sanction Cowork with **Organization IDs**. A Cowork session reports its organization but no plan or email, so a listed organization is accepted only while **Plans** is empty — a Plans list refuses an account that reports no plan.

### Blocked attempts

Every refusal is recorded, so a developer who is stopped is visible to you rather than silent. Open the policy from **Policies → Usage → Account Access** to review its refusals.

Each row carries when it happened, the tool, the Unbound user whose key was used, the account that was refused — its organization ID, email, email domain and plan — the device, and the reason.

Reasons are: *Organization not approved*, *Domain not allowed*, *Plan not allowed*, *Policy from another org*, and *Policy check failed*.

### Accounts in use

Separately from any policy, Unbound records every account a developer signs a coding tool into, so you can see what is in use before you decide what to sanction. Find it under **Analytics → Usage → Account Usage**: the email domains in play across your organization, and a searchable table of accounts with a column each for the user, the tool, the account, its plan and when it was first seen.

> [!WARNING]
>
> Blocked-attempt records contain personal data — a work email address and, where the tool reports it, a hardware serial number. They are kept for as long as your organization exists; there is no automatic cleanup and no configurable retention window today.

> [!NOTE]
>
> Account Access is in **Beta** and may not be enabled for your organization. Policies · Edit creates and changes these policies; Policies · View reads them and their blocked attempts. Each blocked attempt also fires the `account_access.blocked` [webhook](https://docs.getunbound.ai/integrations/webhooks.md#account-access-events). There is no Slack alert or export for a blocked attempt.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.