Is ChatGPT safe for business data? Plus an AI use policy

Whether ChatGPT is safe for business data depends far more on the plan and account than on the brand. This guide compares what OpenAI, Microsoft and Google commit to, gives staff simple data rules, and includes a full AI acceptable-use policy you can adapt.

The short answer

ChatGPT is reasonably safe for business data when staff use a business plan signed in with work accounts, with clear rules about what data goes in. It isn’t safe when staff paste client files into a free personal account, because by default those conversations can be used to improve the model, and your organisation has no control over the account at all.

The same split applies to Microsoft Copilot and Google Gemini. The table below shows where each line falls.

What each vendor commits to

These terms change, so treat the table as a list of questions to put to each vendor, current at November 2026, and confirm the answers in writing before you sign.

For a fuller comparison that includes running models on your own hardware, see Private AI, ChatGPT or Copilot? A decision guide.

What "not used for training" doesn’t cover

A business plan solves the training question. It leaves five others open.

• Privacy law still applies. The OAIC says privacy obligations apply to any personal information put into an AI system, and to output that contains personal information. Using a tool to infer new information about someone counts as collecting it. This is general information, not legal advice.
• Your own permissions. Microsoft 365 Copilot can read anything the user can open in your tenant. If your SharePoint is overshared, Copilot makes that visible. Our Copilot oversharing checklist covers the fix.
• Retention. Prompts and responses are stored. Find out how long, who in your organisation can search them, and whether they’re covered by your retention and eDiscovery settings.
• Web search. When Copilot searches the web, Microsoft sends a short query to Bing with user and tenant identifiers removed, and Bing handles it under separate terms, with Microsoft as an independent controller.
• Accuracy. Every vendor says outputs can be wrong. A human has to check anything that goes to a client, a regulator or a decision about a person.

Simple data rules staff can follow

Most AI policies fail because staff can’t tell, in the moment, whether something is allowed. A four-level rule fixes that.

If your organisation needs AI for confidential or sensitive data that shouldn’t leave your control at all, a private or local AI setup keeps the model on hardware you control.

Where the template comes from

In October 2025 the National AI Centre published the Guidance for AI Adoption. It’s the first update to Australia’s Voluntary AI Safety Standard and condenses its 10 guardrails into six essential practices, often called AI6:

1. Decide who is accountable.
2. Understand impacts and plan accordingly.
3. Measure and manage risks.
4. Share essential information.
5. Test and monitor.
6. Maintain human control.

The Centre also publishes an AI policy guide and template (current version published 22 April 2026) and an AI register template, and suggests using the policy template as a starting point and tailoring it to how your organisation works. The policy below does that. It follows the government template’s structure (purpose, scope, principles, roles, new-use screening, incidents and review) and paraphrases its principles in plainer language, then adds the pieces a small organisation needs day to day: approved tools, data rules, account rules and a staff acknowledgement. For the full government wording and guidance notes, download the original. Have your own legal or HR adviser review your final version.

AI acceptable-use policy template

Replace everything in square brackets.

[Organisation name] AI acceptable-use policy

Version: [number] Approved by: [CEO / board] Date: [date] Next review: [date]

1. Purpose

This policy sets out how [Organisation name] uses artificial intelligence (AI) safely, lawfully and in line with our values. It aims to protect the people whose information we hold, help staff use AI productively, keep our AI use transparent, and give us a clear way to identify and manage AI risks.

2. Scope

This policy applies to all employees, contractors, volunteers and board members who use AI for [Organisation name] work, on any device.

It covers every AI system we use, whether we built it, bought it or got it as a feature inside other software. For this policy, AI means software that works from data to produce content, predictions, recommendations or decisions with some independence from the person using it: chat assistants, AI features in office, email and accounting software, transcription and meeting-summary tools, and AI agents.

Ordinary spreadsheet formulas, fixed if-then automations and standard reports aren’t covered. If you’re not sure about a tool, ask the AI policy owner.

3. Principles

• Human judgment comes first. AI supports our work; people make the decisions and are responsible for them.
• Every AI system has an accountable person. No AI system is used until an owner is named.
• We assess before we adopt. New AI uses are screened for risk before use, with controls that match the risk.
• We are open about AI use. We tell people when they’re dealing with AI or when AI has significantly shaped a decision or document that affects them.
• We protect information. Privacy, confidentiality and security rules apply to AI exactly as they apply to everything else.
• We are fair. AI must not be used in a way that unfairly discriminates against any person or group.

4. Approved tools and accounts

4.1 Only use AI tools listed in our AI register for [Organisation name] work. The register records each tool, its owner, what it’s approved for and the highest data level allowed in it.

4.2 Sign in to approved tools with your [Organisation name] work account only. Don’t use personal AI accounts for work, and don’t use work data in them.

4.3 Where an approved tool has a setting that allows your content to be used for model training, it must be turned off. IT manages this centrally where possible.

4.4 Don’t install AI browser extensions, plug-ins, meeting bots or agents, or connect AI tools to our email, files or systems, without approval from [IT contact].

5. What data you can put into AI tools

Never enter passwords, MFA codes, API keys or other credentials into any AI tool.

6. Using AI outputs

6.1 You’re responsible for anything you produce with AI help. Check facts, figures, names, quotes, references and code before you rely on or share them.

6.2 Don’t present AI output to clients, regulators, funders or the public as verified without a human review.

6.3 Tell the recipient when AI has substantially written a document or response that affects them, or when they’re interacting with an AI system rather than a person.

6.4 AI must not make or substantially make a decision that significantly affects a person (for example hiring, discipline, eligibility, pricing or service access) without approval from the AI governance lead and a documented human review of each decision.

6.5 Check that AI-generated content doesn’t copy protected material or breach someone else’s rights before you publish it.

7. Prohibited uses

Don’t use AI to:

• deceive, manipulate or impersonate anyone, including creating fake reviews, testimonials or voices
• monitor or profile staff, clients or members of the public without approval and a privacy assessment
• produce unlawful, discriminatory, harassing or explicit content
• get around a security control, licence or another policy.

8. New AI tools and uses

8.1 To request a new AI tool or a new use of an existing one, complete [request form] and send it to [AI policy owner].

8.2 Each request is screened and given an outcome: normal (approve and add to the register), elevated (needs a risk and privacy assessment and approval by [AI governance lead or committee]) or prohibited.

8.3 Requests involving personal or sensitive information, decisions about people, customer-facing AI or AI agents that can take actions are always at least elevated.

9. Roles and responsibilities

10. Reporting problems

Report immediately to [contact] if you:

• entered data into an AI tool that this policy doesn’t allow
• notice AI output that is harmful, discriminatory, or exposes information it shouldn’t
• suspect an AI tool or account has been misused or compromised.

Incidents are handled under our [incident response plan]. If personal information may be involved, our privacy officer assesses whether the Notifiable Data Breaches scheme applies. We’ll keep the ability to switch off any AI system and fall back to a manual process when needed.

11. Training

Everyone covered by this policy completes AI awareness training when they start and when this policy changes significantly.

12. Review

This policy is reviewed [review period] by the AI policy owner, and sooner after a significant AI-related incident, a major change in the AI tools we use, or a change in relevant law or standards. Changes are approved by the policy approver.

13. Acknowledgement

I have read and understood the [Organisation name] AI acceptable-use policy and agree to follow it.

Name: ________ Signature: ________ Date: ________

Structure and principles adapted from the National AI Centre’s AI policy guide and template (Department of Industry, Science and Resources), part of the Guidance for AI Adoption. © Commonwealth of Australia.

Putting it in place

1. Find out what’s already in use. Ask teams which AI tools they use and for what. Expect more than you think, including AI features switched on inside existing software.
2. Pick the approved tools and turn off training and unnecessary connectors in each admin console.
3. Start the AI register. One row per tool: owner, approved uses, data level, settings checked. The National AI Centre publishes a register template.
4. Adapt and approve the policy above, then brief staff with the data table rather than the whole document.
5. Block or discourage what’s not approved, and give people a quick way to request new tools so they don’t go around you.

You can see how we describe our own use of AI in our AI use policy. If you’d like help with the register, risk screening or staff training, see AI security and governance.

Key takeaways

• The plan and account decide whether your data trains a model: personal accounts generally can, business accounts generally don’t.
• Business terms solve training, not privacy, permissions, retention or accuracy. Those need your own rules.
• Follow the OAIC: keep personal and sensitive information out of publicly available AI tools.
• A four-level data table is the part of an AI policy staff actually remember.
• The government’s six AI practices and policy template are a sound base; the adapted policy above turns them into rules staff can follow.

Sources

• OpenAI, Business data privacy, security and compliance and Enterprise privacy
• OpenAI Help Center, Data Controls FAQ
• OpenAI Help Center, Data residency for ChatGPT
• OpenAI API documentation, Data controls and data residency
• Microsoft Learn, Enterprise data protection in Microsoft Copilot and Copilot Chat
• Microsoft Learn, Data, privacy and security for Microsoft Copilot
• Microsoft, In-country data processing for Microsoft 365 Copilot (editor’s note, 3 April 2026)
• Microsoft Support, Privacy FAQ for Microsoft Copilot
• Google, Generative AI in Google Workspace Privacy Hub
• Google, Gemini Apps Privacy Hub
• OAIC, Guidance on privacy and the use of commercially available AI products
• National AI Centre, Guidance for AI Adoption: Foundations (October 2025)
• National AI Centre, Create an AI policy: AI policy guide and template (published 22 April 2026)