[Guides](https://muteware.com/guides/)

# Temporary chat, a business plan, a local model or scrubbing first: what each one protects

By Muteware15 November 20263 min read

In this guide

1. [Three moments where things can go wrong](https://muteware.com/guides/temporary-chat-vs-business-plan-vs-scrubbing/#three-moments-where-things-can-go-wrong)
2. [Side by side](https://muteware.com/guides/temporary-chat-vs-business-plan-vs-scrubbing/#side-by-side)
3. [Settings and contracts work after the text has left](https://muteware.com/guides/temporary-chat-vs-business-plan-vs-scrubbing/#settings-and-contracts-work-after-the-text-has-left)
4. [Scrubbing works before the text leaves](https://muteware.com/guides/temporary-chat-vs-business-plan-vs-scrubbing/#scrubbing-works-before-the-text-leaves)
5. [What to combine](https://muteware.com/guides/temporary-chat-vs-business-plan-vs-scrubbing/#what-to-combine)
6. [Tooling](https://muteware.com/guides/temporary-chat-vs-business-plan-vs-scrubbing/#tooling)

There are four common answers to the question of how to use an AI chat without leaking personal data or secrets: use a temporary chat, buy a business plan, run a model on your own computer, or remove the sensitive parts before you paste. They are often presented as alternatives. They are not. Each one closes a different door.

## Three moments where things can go wrong

1. **Before sending.** You paste something you were not supposed to share at all: a customer list, a key, a medical detail.
2. **At the provider.** Your text is stored, may be read during abuse checks, and may be used to improve models, depending on the plan and your settings.
3. **Afterwards.** The conversation stays in a history that others with access to the account or the device can read.

## Side by side

| Measure | What it changes | What it leaves open |
| --- | --- | --- |
| Temporary or private chat | The conversation does not stay in your history and, as providers describe it, is not used to train models | The text still goes to the provider and is processed there. A copy can be kept for a limited time for safety reasons |
| Switching off training in the settings | Your conversations are no longer used to improve the models | They are still sent, stored and kept in your history |
| Business or enterprise plan | A contract with a data processing agreement, no training on your data by default as providers state for these plans, and controls for an administrator | The data still leaves your organisation, and nothing stops a colleague from pasting what should not be shared |
| A model on your own computer | Nothing leaves the machine | You need the hardware and the know how, and local models are usually less capable than the large hosted ones |
| Scrubbing before you paste | The sensitive values never leave your computer, whatever tool, plan or setting is on the other side | It finds patterns, not meaning. Names and context have to be added or checked by you |

## Settings and contracts work after the text has left

The first three measures are promises by the provider about what happens to your text once they have it. Those promises matter, and for work you want them in a contract. But they do not change the fact that the text was sent. If the text held something you had no right to share, such as personal data of customers without a legal basis, a better setting does not repair that.

Names and defaults of these settings change regularly. Check the current help pages of the tool you use. For ChatGPT, OpenAI describes them in its [data controls FAQ](https://help.openai.com/en/articles/7730893-data-controls-faq).

## Scrubbing works before the text leaves

Scrubbing is the only one of the four that acts at the first moment. You replace email addresses, phone numbers, account numbers and keys with placeholders such as \[EMAIL\_1\], send the text, and put the real values back in the answer. The AI still understands the task, because the structure of the text is intact. The [placeholder method](https://muteware.com/guides/placeholder-method/) explains why this works so well.

Its limit is just as clear. A scrubber recognises things that have a shape. It does not know that "the director of the school in our village" points to one person. Add names and project words as your own terms, and read the result before you send it.

## What to combine

- **On your own, with a free or personal plan:** switch off training, use a temporary chat for anything you do not need to keep, and scrub what is sensitive.
- **In a company:** a business plan with a data processing agreement, a short [AI policy](https://muteware.com/ai-policy-template/) that says what never goes in, and scrubbing as the habit that makes the policy easy to follow.
- **For the most sensitive material:** do not send it. Use a local model, or leave AI out of it.

## Tooling

[Scrub for AI](https://muteware.com/scrub-for-ai/) does the scrubbing and the restoring in your browser, and you can try the same engine online at [muteware.com/scrub](https://muteware.com/scrub/) without installing anything.

This guide is general information and not legal advice.

### Scrub before you paste

Scrub for AI replaces emails, phone numbers, IBANs, API keys and passwords with placeholders in one click, and puts them back when the answer returns. Free, runs offline, no account.

[Add to Chrome, free](https://chromewebstore.google.com/detail/scrub-for-ai-hide-persona/madopcokggkmjlmbiifobacolinpabbg) [Firefox](https://addons.mozilla.org/firefox/addon/scrub-for-ai/) [Try it online](https://muteware.com/scrub/)

More: [all guides](https://muteware.com/guides/) · [Scrub for AI](https://muteware.com/scrub-for-ai/) · [Copy for AI](https://muteware.com/copy-for-ai/) · [Tables for AI](https://muteware.com/tables-for-ai/)

---

Source: https://muteware.com/guides/temporary-chat-vs-business-plan-vs-scrubbing/
Summary: Four ways to keep personal data and secrets safe when you use AI chats, compared by what they change and what they leave open. They work best together.
Publisher: Muteware, support@muteware.com
