Secure AI for Business

Is it safe to put company data into ChatGPT or Claude?

The answer depends less on the assistant than on the plan. Here is what OpenAI, Anthropic, Microsoft and Google say about company data on each, and what stays the company's job.

Contents
  1. The plan matters more than the brand
  2. What a business plan does and does not settle
  3. Copilot and Gemini: the same question inside the suite
  4. What to do on Monday
  5. Common questions
  6. Sources

The question usually arrives after the fact: a sales manager has pasted a customer's contract into a chatbot to summarise it, or a price list to reformat it. Whether that was a problem depends on which account they used.

The plan matters more than the brand

ChatGPT and Claude each come in consumer and business versions with different data terms. Staff who sign up on their own almost always have the consumer version.

PlanTraining on conversationsSource
ChatGPT Free, Go, Plus, ProOpt-out availableOpenAI pricing page⁠1
ChatGPT Business, Enterprise"Your data is never used for training our models"OpenAI Business page⁠2
Claude Free, Pro, MaxTrained on when the user's setting is on; retention up to five years if soAnthropic, 28 August 2025⁠3
Claude Team, Enterprise, APINone by defaultAnthropic pricing page and privacy center⁠45
Microsoft 365 Copilot and Copilot Chat (work account)Not used to train foundation modelsMicrosoft Learn⁠6
Gemini in Google WorkspaceNot used for training outside the domain without permissionGoogle Workspace privacy hub⁠7

So the short answer is: a business plan, approved by the company, removes the training risk by default. A personal account may not.

What a business plan does and does not settle

What it settles

  • Training. All four vendors exclude business data from model training by default⁠2567.
  • Account control. ChatGPT Business includes SSO, multi-factor authentication and admin controls for users, usage and billing⁠2; Claude Team includes SSO⁠4.
  • Encryption. OpenAI encrypts data in transit with TLS 1.2 and at rest with AES-256⁠1. Microsoft encrypts Copilot data at rest and in transit and isolates it between tenants⁠6.

What it leaves to the company

  • Retention and audit. On Claude, audit logs and custom retention controls come with Enterprise, not Team⁠4. On ChatGPT, SCIM, role-based access and data residency go through sales⁠2.
  • Feedback. Anthropic may use thumbs-up or thumbs-down feedback for training after de-linking it from user and customer IDs⁠5.
  • Who sees what inside the company. Assistants connected to company files show each user what that user can already open⁠87.
  • What the AI may do. When an assistant can act, sending, changing or paying, the risk is no longer only disclosure. OWASP recommends least-privilege access and human approval for high-risk actions⁠9.

Copilot and Gemini: the same question inside the suite

For companies already on Microsoft 365 or Google Workspace, the assistant inside the suite is often the simplest answer, because the data never leaves the contract the company already has. Microsoft covers Copilot prompts and responses under the same terms as its email and files and applies the company's retention policies⁠6. Google says Gemini interactions stay within the organisation and that existing Workspace protections apply automatically⁠7. Both retrieve only what the user can access, which is the strength and the condition.

What to do on Monday

  1. Find out which accounts staff use. Ask directly; most will say. In Microsoft's 2024 survey, 78% of AI users brought their own tools to work⁠10.
  2. Approve one business plan and move work onto it. Choose the one inside the suite the company already pays for, unless staff depend on a specific assistant.
  3. Publish the never-list: customer personal data, NDA material, prices and margins, personnel files.
  4. Set the enterprise controls you need: SSO, retention, audit.
  5. Decide what AI may do before it does it. For any AI that acts, write who approves. Some systems make this a setting: OpenLazarus lets a company set data, files, messages and spending to Always, Ask or Never, give a teammate a credit card with spending permissions, and decide who approves, with every action and approval logged⁠11.

The end state is simple to describe. Company data goes only into the approved tool, under business terms that exclude training. Staff know what never goes in. And the owner can answer, on any given day, where prompts are stored and who can read them. The data security checklist covers the remaining checks.

Common questions

Does ChatGPT train on company data?

On ChatGPT Business and Enterprise, OpenAI says business data is never used to train its models. On the Free, Go, Plus and Pro plans, training is on unless the user opts out, according to OpenAI’s pricing page.

Does Claude train on company data?

Not by default on Team, Enterprise or the API. On consumer Free, Pro and Max accounts, Anthropic trains on chats when the user’s setting is on and keeps that data for up to five years; otherwise it keeps the existing 30-day period.

Is it safe to put customer data into ChatGPT?

Only on a business plan approved by the company, and only for data the company has decided may go into AI. A business contract stops training on the data; it does not decide who in the company should see it or how long prompts are kept.

Is Microsoft Copilot safer than ChatGPT for company data?

Not inherently. Copilot works inside Microsoft 365 under the company’s existing contract, permissions and retention policies, which suits companies already on Microsoft. It also surfaces anything a user can open, so it is only as safe as the company’s sharing settings.

Sources

  1. ChatGPT pricing. OpenAI. Accessed 19 Nov 2026.
  2. ChatGPT Business. OpenAI. Accessed 19 Nov 2026.
  3. Updates to consumer terms and privacy policy. Anthropic, 28 Aug 2025. Accessed 19 Nov 2026.
  4. Plans and pricing. Anthropic. Accessed 19 Nov 2026.
  5. Is my data used for model training?. Anthropic Privacy Center, 18 Aug 2026. Accessed 19 Nov 2026.
  6. Enterprise data protection in Microsoft Copilot and Microsoft Copilot Chat. Microsoft Learn, 29 May 2026. Accessed 19 Nov 2026.
  7. Generative AI in Google Workspace Privacy Hub. Google. Accessed 19 Nov 2026.
  8. Data, privacy, and security for Microsoft Copilot. Microsoft Learn, 9 Jul 2026. Accessed 19 Nov 2026.
  9. LLM01:2025 Prompt Injection. OWASP Gen AI Security Project. Accessed 19 Nov 2026.
  10. AI at work is here. Now comes the hard part (2024 Work Trend Index). Microsoft and LinkedIn, 8 May 2024. Accessed 19 Nov 2026.
  11. OpenLazarus. OpenLazarus. Accessed 19 Nov 2026.

Updated 19 Nov 2026: First published, with data terms read on each vendor’s pages.