---
title: "How to use AI while protecting customer data, trade secrets, and company know-how"
description: "5 source-backed AI claims from How to use AI while protecting customer data, trade secrets, and company know-how, with key statistics, context, and the…"
url: "https://kbanc.com/claims-library/how-to-use-ai-while-protecting-customer-data-trade-secrets-and-company-know-how"
source: "https://aiadopters.club/p/before-you-upload-that-spreadsheet"
date: "2026-09-30"
topics: ["business", "implementation", "strategy"]
generated: "2026-10-02"
---

# How to use AI while protecting customer data, trade secrets, and company know-how

By Kamil Banc | September 30, 2026

## Claims

1. **Anthropic's Default Training Policy** (source summary): Anthropic's commercial products exclude customer data from model training by default, with exceptions including explicit opt-ins.
2. **Client Consent at Morgan Stanley** (source summary): Morgan Stanley's Debrief tool uses client consent before AI generates meeting notes and action items.
3. **Mercedes-Benz Employee Usage Rules** (source summary): Mercedes-Benz guidance tells employees to avoid personal or internal data in inputs to Direct Chat.
4. **Prompting Cannot Unsend Data** (direct quote): Writing 'ignore the confidential columns' in a prompt has already sent those columns to the model.
5. **Four Separate AI Controls** (Kamil's interpretation): A no-training promise does not settle storage, access, or what connected tools can do.

## Evidence

### Quote
> "A competitor wouldn't need the names to learn quite a lot about your business." - Kamil Banc

### Key Statistics
- **4**: Separate controls the author says businesses must evaluate when choosing an AI tool: training, storage, access, and actions.
- **8**: Number of shares the article received on Substack as of its September 30, 2026 publication date.
- **1**: Number of recurring tasks the author recommends starting with this week when mapping where sensitive data flows.

## Context
The author, a practitioner who coaches business owners on AI adoption, grounds his guidance in company-described practices from Mercedes-Benz and Morgan Stanley, alongside documented policies from Anthropic on training and data retention. His methodology treats data protection as four independent controls—training, storage, access, and actions—rather than a single yes-or-no question about model training. For application, he recommends a workflow where business software strips unnecessary fields before data reaches the model, the model drafts output, and a human reviews and approves any real action. He also advises beginning with fictional data in one recurring task, tracing where information appears in logs and connected tools, and only then deciding what additional access the results justify. The framework is practitioner-oriented rather than academic, but each recommendation ties back to verifiable company practices or documented product policies.

## Source
- Original: [How to use AI while protecting customer data, trade secrets, and company know-how](https://aiadopters.club/p/before-you-upload-that-spreadsheet)
- Cite: kbanc.com/claims-library/how-to-use-ai-while-protecting-customer-data-trade-secrets-and-company-know-how

## Primary Evidence
- [Anthropic’s commercial products exclude training by default](https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training) (privacy.claude.com; supports claim 1)
- [Debrief launch announcement](https://www.morganstanley.com/press-releases/ai-at-morgan-stanley-debrief-launch) (morganstanley.com; supports claim 2)