# Personal information masking for AI

Privacy replaces detected personal information with temporary placeholders for selected AI requests and restores it in supported replies and tool inputs.

*Release 3.3 · admin, operator, visitor*

**Settings › AI › Accounts** · **Settings › AI › Connections**

## What it is

**Privacy** can replace detected personal information with placeholders before selected AI requests. INSIDE keeps a reversible mapping so supported answers and tool inputs can use the original values afterwards. This reduces the personal information in those model requests while keeping the conversation usable.

## Who uses it and why

Administrators ask Powerfront to select privacy processing for the relevant AI accounts. Operators and visitors continue using normal conversation text; they do not manage the placeholders.

## Where to find it

Powerfront manages **Privacy** in **Settings › AI › Accounts** and the associated service in **Settings › AI › Connections**. Ask Powerfront to configure or change them for you.

## How it works

1. On a supported AI flow, INSIDE analyses visitor and operator messages for personal information.
2. Detected values receive placeholders that can be reused across the conversation.
3. Before generation, INSIDE replaces known values in instructions, message text, tool inputs and tool results.
4. After generation, INSIDE restores recognised placeholders in answer text, tool inputs and supported rich content.

Supported categories include names, locations, email addresses, telephone numbers, IP addresses, credit card numbers, IBANs, US social security numbers and US passport numbers. Detection depends on the analyser and the text it receives.

The original conversation and the reversible mapping can still contain personal information. This feature does not permanently erase it.

## Configure it

Ask Powerfront to prepare the privacy service and select it in **Privacy** for each applicable AI account. The initial selection is **None**. An account must have a privacy service selected and privacy processing enabled for supported operations to use it.

Agree which AI features use that account, then verify representative messages and replies. A successful configuration test alone does not prove that the analyser detects your expected personal information.

### Settings

| Setting | Where | Applies to | Default | Who can change it | What it does |
|---|---|---|---|---|---|
| Privacy | Settings › AI › Accounts | website | None | Powerfront, on request | Selects the optional privacy connection used by supported AI operations on this account. |

## Limits and good to know

- Detection and replacement are best effort. An analyser or processing failure can allow work to continue without replacement.
- Known-value replacement in a tool result does not mean every new personal value in that result is analysed.
- Do not assume coverage of attachments, images or live audio.
- Language detection for translation excludes names and locations from replacement.
- Original values can reappear in replies and tool inputs after restoration. Placeholders do not prevent authorised tools from receiving those values.
- Privacy processing is separate from message masking, access restrictions and data retention.

## Depends on and works with

### This feature uses

- **Requires** [Shared AI service connections](AIP-013.md): Automatic privacy processing needs a selected privacy service and an available analyser.
- **Configured in** [AI accounts and shared defaults](AIP-001.md): The AI account selects the privacy connection and controls whether privacy processing is enabled.
- **Works better with** [AI conversation summaries](AI-009.md): Supported summaries and metrics use optional anonymisation when the account has privacy enabled and selected.
- **Works better with** [Chat Translate and language detection](AI-010.md): Supported language and translation operations can use privacy processing with operation-specific exclusions.
- **Works better with** [Rich content in AI replies](AI-026.md): Restoration includes supported text within rich content and reply options.

### Used by

- **Enhances** [AI agent service connections](AI-024.md): Configured personal-data protection also applies to external execution and tool results.
- **Enhances** [AI accounts and shared defaults](AIP-001.md): A privacy connection can anonymise AI inputs and restore protected values in replies.
- **Enhances** [Shared AI service connections](AIP-013.md): Privacy connections support sensitive-data processing.
- **Enhances** [AI execution tracing with Langfuse](AIP-015.md): Configured privacy processing can replace known personal values in supported trace inputs without guaranteeing complete redaction.

## FAQ and troubleshooting

### Why did personal information reach an AI request?

Ask Powerfront to check the selected **Privacy** connection, account enablement, analyser availability and whether the operation supports this processing. Detection can also miss a value.

### Why does the answer still show the person's name?

INSIDE restores recognised placeholders in supported answers. The purpose is to reduce personal information in selected model requests while preserving useful replies.

### Does selecting Privacy erase stored personal information?

No. The conversation and reversible mapping can retain original values. Your separate access and retention settings still matter.

### Does this cover spoken calls and uploaded documents?

Do not assume that it does. Ask Powerfront to confirm coverage for the exact operation and content type you use.