Data, handled carefully

Is it safe to use ChatGPT with company data?

Using ChatGPT with company data can be acceptable when the organisation has chosen a plan whose data terms it has read and approved, and staff are trained on what may and may not be entered, but pasting confidential or personal data into a personal account is not safe practice.

By Toshendra · Founder

Is it safe to use ChatGPT with company data, in short?

It depends on three things: which account or plan is used, what that plan's data terms say, and what information people put in. The same tool can be acceptable on an approved business plan for a draft built from public facts, and unacceptable on a personal account for a client contract.

The UK National Cyber Security Centre made the underlying point in March 2023: a query is visible to the organisation providing the model, and it recommended not including sensitive information in queries to public models. Vendor terms have developed since, but the principle that you must know where the data goes still holds. This page is general information, not legal advice.

Choose the plan

Decide which account type is approved for work. Personal accounts are a common gap.

Read the terms

Check training use, retention, admin controls and where data is processed, on the vendor's own pages.

Classify information

Agree what is public, internal, confidential and personal, with simple examples.

Set the rule

Write a short list of what must never be entered, whatever the plan.

Train and check

Practise with fictional data, and review real use against the rule.

Does the type of plan change what happens to your data?

Yes, vendors generally publish different data terms for consumer and business products, so check the exact plan you use. Anthropic's privacy centre, dated 18 August 2026, says that by default it will not use inputs or outputs from its commercial products to train its models, and points consumer plans to a separate article. Google's Workspace privacy hub says Workspace does not use customer data for training models without the customer's prior permission or instruction.

OpenAI's API documentation says data sent to the API is not used to train or improve its models unless you opt in, and that abuse monitoring logs are retained for up to 30 days by default. We could not open OpenAI's ChatGPT business and consumer privacy pages when we checked on 16 October 2026, so read those directly for the ChatGPT plan your team uses rather than relying on a summary here.

What should employees never paste into ChatGPT?

Some information is a poor fit for any general AI tool unless a specific, approved process covers it. A training default that works for most small teams is a short never-enter list, with an owner who answers edge cases.

Passwords and keys

Credentials, API keys, access codes and security details.

Special category personal data

Health, biometric and similar sensitive details about anyone.

Client confidential material

Anything a contract or NDA restricts, unless the client has agreed.

Unreleased financials

Results, deal terms or pricing not yet public.

Personnel matters

Disciplinary, grievance or performance details about named staff.

Want to see how this connects to a role?

Explore the planned team trial ↗

What does a safer prompt look like?

At Copperfield Consulting, a fictional 20-person firm, a consultant wants help summarising a client interview. The risky version pastes the full transcript, including the client's name, a named employee's complaint and unreleased revenue figures, into a personal account.

The safer version uses the firm's approved business plan, removes the names and figures, and asks: "Summarise the main themes from these anonymised notes in five bullet points. Do not infer anything not stated." The consultant then checks each bullet against the notes. The firm's policy states which plan is approved and that client data needs the engagement lead's agreement.

What about personal data and data protection law?

If staff enter personal data, your data protection obligations apply in the usual way, and AI does not remove them. NIST's Generative AI Profile lists data privacy, including leakage and unauthorised disclosure of personal or sensitive data, as a generative AI risk. In the UK, the ICO publishes guidance on AI and data protection; its summary page, dated 15 March 2023, says it is under review following the Data (Use and Access) Act. Take advice for anything high-risk.

Where does Tosheo.ai fit?

Tosheo.ai is being built to teach information boundaries as part of practical lessons, using fictional or approved material, so people practise the decision before they face it with real data. Planned product commitments are that customer documents are isolated per company, deleted on account closure and never used for training. These are commitments for a pre-build product, not audited controls.

When is Tosheo.ai not the answer?

Training does not choose your AI plan, negotiate a data processing agreement or carry out a risk assessment. If you need a security review, a data protection impact assessment or contract terms, work with your IT, legal or data protection lead. Tosheo.ai does not offer SSO or enterprise contracts at launch.

Sources and scope

UK NCSC, ChatGPT and large language models: what's the risk?, 14 March 2023 ↗

Anthropic Privacy Center, Is my data used for model training?, dated 18 August 2026 ↗

Google Workspace, Generative AI in Google Workspace Privacy Hub, checked 16 October 2026 ↗

OpenAI API documentation, Your data, checked 16 October 2026 ↗

NIST AI 600-1, Generative AI Profile, July 2024 ↗

ICO, Guidance on AI and data protection, 15 March 2023, under review ↗

Before you decide

A few good questions.

Does ChatGPT train on what I type?

It depends on the plan and settings. Vendors publish different terms for consumer and business products, and some offer controls. Read the current data page for the exact ChatGPT plan your team uses.

Is a business plan automatically safe?

No. Business terms can reduce some risks, such as training use, but staff can still paste in information a contract or law restricts. You still need a clear rule on what may be entered.

Can we ban ChatGPT instead?

You can, but people may use personal accounts anyway, which is harder to see. An approved tool with clear rules and training is often easier to manage than a ban nobody follows.

Is anonymising data enough?

Removing names helps, but details can still identify a person or client in context. Remove identifying details where you can, and keep restricted material out unless an approved process covers it.

Who should own AI data rules in a small company?

Name one person, often the operations lead or the person responsible for data protection, to answer edge cases and keep the rule current as tools and plans change.

Give your team a path, not another playlist.

Explore the planned role-specific lessons, practical tasks and manager view.

Start free trial7 days free · Card required · Up to 5 seats