ChatGPT at Work: How to Use It Safely Without Risking Your Data

Published 4 July 2026 · 7 min read · By Richard Brereton

ChatGPT has become one of the most widely used workplace tools of the past three years. It drafts, summarises, translates, codes, and answers questions faster than any human could. The problem is that most people using it at work are doing so in ways that create real data, legal, and reputational risk — without knowing it.

This guide covers how to use ChatGPT productively at work while protecting your organisation from the most common mistakes.

The Core Risk: What You Paste Goes to OpenAI

Every time you paste text into ChatGPT, that text is transmitted to OpenAI's servers. Under default settings, OpenAI may use your conversations to train future models. This means that confidential company information, client data, employee records, financial figures, and legal documents could all end up in OpenAI's training pipeline.

This is not theoretical. In 2023, Samsung employees accidentally leaked semiconductor schematics and internal meeting notes through ChatGPT. Samsung subsequently banned the tool internally. The data was already sent before the ban.

What Should Never Go Into ChatGPT

• Employee personal data — names, salaries, NI numbers, performance reviews, health information
• Client or customer records — contact details, account information, purchase history
• Legal documents — contracts, NDAs, privileged communications
• Financial data — bank details, card numbers, unreported results
• Intellectual property — source code, product roadmaps, trade secrets
• Strategic plans — M&A discussions, pricing strategies, competitive intelligence

Safe Ways to Use ChatGPT at Work

Anonymise before you paste

The most effective habit is to remove or replace all personal and confidential data before submitting anything to ChatGPT. Replace real names with placeholders, remove specific financial figures, and strip out email addresses and phone numbers. Then paste the anonymised version.

Doing this manually is slow and unreliable. Mutant Data Safety Layer automates the process — paste your text, it scans for personal data and replaces it with typed placeholders like [EMAIL_REDACTED] or [NI_NUMBER_REDACTED], and you copy the clean version into ChatGPT.

Turn off training on your account

In ChatGPT, go to Settings → Data Controls → Improve the model for everyone and turn it off. This reduces (but does not eliminate) the risk of your data being used for training.

Use the right tier for sensitive work

ChatGPT Enterprise includes a Data Processing Agreement and does not use your data for training by default. If your organisation handles large volumes of sensitive data and AI use is essential, Enterprise is the appropriate tier. Free and Plus accounts have no DPA and should not be used for sensitive business data.

Keep a clear approved-use policy

Organisations should have a written policy covering which AI tools are approved, what categories of data can be shared, and what the process is for checking documents before submission. Without a policy, individual employees make their own judgements — and those judgements can have serious consequences.

Getting the Most Out of ChatGPT Safely

None of this means ChatGPT isn't valuable at work. It is. The key is structuring how you use it:

• Use it for drafting, structuring, and rewriting — not for processing raw personal data
• Feed it anonymised examples, not real records
• Use it to generate templates that you then personalise manually
• Use it for research, summarisation of public information, and ideation

These use cases are high-value and carry minimal risk. The problems arise when people start pasting in entire client emails, HR records, or financial documents without thinking about what's in them.