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

# PII Enforcement

> Mask personal data in every message before it reaches an AI model

PII enforcement makes sure personal data, such as names, email addresses and phone numbers, never reaches an AI provider in plain text. It applies to every member of your organization and can't be switched off by individual members.

## Turn on PII enforcement

PII enforcement is set per organization by BlackMountain. To turn it on or off, contact BlackMountain support.

Once it is on, members see **Personal Information Encryption activated** in the chat.

## How it works

1. **Detect.** Before a message is sent, 48° scans the text and any attached files for personal data.
2. **Review.** The member sees **Confirm personal information before sending**, with each finding highlighted. They can remove a mask, change its type, or select other text and mark it as personal information. Then they click **Confirm & send**.
3. **Mask.** Each confirmed finding is replaced with an encrypted token. Only the masked text is saved and sent to the model.
4. **Reveal on display.** The model answers using the tokens. The member who owns the chat can click **Show original personal information** to see the original values again in their own browser.

The original values are stored encrypted, separately from the chat. The model and the AI provider only ever see the masked text.

## What is detected

| Type | Examples |
| - | - |
| **Person** | Names of people |
| **Email** | Email addresses |
| **Phone number** | Phone numbers |
| **Location** | Cities, addresses, places |
| **IBAN** | Bank account numbers |
| **Credit card** | Card numbers |
| **US SSN** | US social security numbers |
| **IP address** | IPv4 and IPv6 addresses |

Members can also mark any other text as **Custom**. Detection works for English and German text.

> [!WARNING]
>
> Automatic detection is not perfect. It can miss personal data or mark text that isn't personal. Ask members to check the highlights in the review dialog before they send.

## If the protection service is unavailable

PII enforcement fails safe. If the text can't be checked, the message is not sent. The member sees **PII redaction service is unavailable. Your organization requires PII protection — please try again shortly.** No unchecked text is sent to a model.

## What changes for members

While PII enforcement is on:

* **Chat sharing is unavailable.** Members see **Sharing is unavailable because your organization enforces PII protection.**
* **Memory is unavailable.** The assistant does not remember facts across chats.
* **Assistant instructions and knowledge files are not scanned.** They reach the model as written. The Assistant Studio shows a warning for both. Keep personal data out of them.
* **Editor documents** are stored as written and masked when the AI reads them.

## Related

> [card] [PII protection for members](https://docs.blackmountain.io/chat/pii-protection.md)
>
> What members see in chat.

> [card] [Why is text masked?](https://docs.blackmountain.io/faq/masked-text.md)
>
> Answers for members.

> [card] [Data residency](https://docs.blackmountain.io/admin/security/data-residency.md)
>
> Where data is stored and processed.

> [card] [Sharing controls](https://docs.blackmountain.io/admin/security/sharing-controls.md)
>
> Control what members can share.


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