AI data safety
Staff are already using ChatGPT with company data. What should our AI use policy say?
Last updated
An AI use policy should name the approved tools and business plans staff may use, list the data staff must never paste into AI tools, require a person to check every output, and say who approves new tools. Start from the Australian Government AI policy guide and template on ai.gov.au, then train staff on it.
How to write an AI use policy staff will follow
Find out what is already in use
Run a short, no-blame survey and check web gateway and single sign-on logs for AI sites and browser extensions. The aim is a list of real uses, not a list of names.
Approve tools on business plans
Give staff at least one sanctioned assistant on a business or enterprise plan, signed in with work accounts. OpenAI states that ChatGPT Business and Enterprise data is not used for training by default, which is not the case for free, Plus and Pro accounts.
Set data classes in plain words
Tie the policy to your existing information classification and give examples staff recognise: a customer complaint email, a payslip, a tender price, a password.
Require human review
Every AI output is checked by the person who uses it before it reaches a customer, a system of record or a decision about an individual. Staff stay accountable for what they send.
Create an approval path for new tools
One short form for any new AI tool or plug-in, reviewed by IT and the privacy lead against data location, training terms and access controls. Record approvals in an AI register.
Train, then review
Run a short session with worked examples for each team, add AI to induction, and review the policy when tools, plans or the law change.
What staff can put into which tool
| Data class | Approved business AI tool | Free or personal AI account |
|---|---|---|
| Public information, such as published web content | Allowed | Allowed |
| Internal, non-sensitive work, such as drafts without names | Allowed | Not allowed |
| Commercial in confidence, such as pricing, tenders, contracts, board papers | Allowed only if the tool is approved for that class | Never |
| Customer or staff personal information | Only for a use case the privacy lead has approved | Never |
| Sensitive information, such as health records, tax file numbers, biometrics | Never, unless a specific use case is approved and assessed | Never |
| Passwords, keys and security settings | Never | Never |
Australian specifics
The National AI Centre publishes an AI policy guide and template, linked from the Create an AI policy page on ai.gov.au, with an editable Word template. It covers purpose, scope, policy statements, governance and review, and is written to be adapted to each organisation's obligations. It is a sound starting structure; your data classes, approved tools and approval path are what make it usable.
The OAIC recommends, as best practice, that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools. Pasting customer details into a consumer account can also raise APP 6 (use and disclosure), APP 8 (overseas disclosure) and APP 11 (security). From 10 December 2026, APP 1.7 to 1.9 require privacy policies to explain certain automated decisions that use personal information, so the policy should require any AI used in decisions about people to be registered and reviewed by the privacy lead.
New Zealand organisations should read the Privacy Commissioner's guidance on AI and the information privacy principles, which recommends a privacy impact assessment before staff use AI tools with personal information.
Limits and when not to use AI
A ban on its own pushes use onto phones
Blocking free tools without offering an approved one moves the same behaviour onto personal devices, where IT has no visibility.
A policy does not enforce itself
Back it with single sign-on, data loss prevention where available, and blocking of unapproved tools on work devices once an approved tool exists.
Keep AI out of final decisions about people
Hiring, discipline, credit and eligibility decisions need a person to decide, with AI limited to drafting or summarising.
Do not use AI where output cannot be checked
Legal advice, safety-critical instructions and figures nobody can verify should not rely on AI output.
Who should own it
Privacy or risk lead
Owns the policy, the data classes and approval of any use involving personal information.
IT manager
Owns the approved tool list, sign-in and blocking controls, and the AI register.
HR lead
Builds the policy into induction, training and the code of conduct.
Executive sponsor
Approves the policy and settles disputes between speed and risk.
Related work
- Shadow AI is already in your businessInventory the AI tools staff use and write a policy that holds.
- ChatGPT and OpenAI consultingA governed ChatGPT rollout with policy, plans and custom GPTs.
- Copilot readiness assessmentCheck what Microsoft 365 Copilot would surface before rollout.
- AI data sovereignty in AustraliaAPP 8 and hosting rules for regulated industries.
Sources
Australian Government, ai.gov.au, Create an AI policy: https://www.ai.gov.au/staying-safe-and-responsible/essential-ai-practices/create-ai-policy
Department of Industry, Science and Resources, AI policy guide and template: https://www.industry.gov.au/publications/guidance-for-ai-adoption/ai-policy-guide-and-template
Office of the Australian Information Commissioner, Guidance on privacy and the use of commercially available AI products: https://www.oaic.gov.au/privacy/privacy-guidance-for-organisations-and-government-agencies/guidance-on-privacy-and-the-use-of-commercially-available-ai-products
Office of the Australian Information Commissioner, New resources on transparency for use of AI and automated decision making: https://www.oaic.gov.au/news/media-centre/new-resources-on-transparency-for-use-of-ai-and-automated-decision-making
Office of the Privacy Commissioner (New Zealand), Artificial intelligence: https://www.privacy.org.nz/resources-and-learning/a-z-topics/ai/
OpenAI, Enterprise privacy: https://openai.com/enterprise-privacy/
Questions
Is there an Australian AI policy template organisations can start from?
Yes. The National AI Centre publishes an AI policy guide and template, linked from the Create an AI policy page on ai.gov.au and available as an editable Word document. It sets out purpose, scope, policy statements, governance and review. Organisations still need to add their own data classes, approved tools and approval process for it to work in practice.
Are free ChatGPT accounts trained on what staff type?
By default, yes. OpenAI states that conversations on free, Plus and Pro accounts may be used to improve its models unless the user turns this off, while business plans such as ChatGPT Team and Enterprise are excluded by default. An AI use policy should therefore keep company data out of personal accounts and give staff an approved business plan instead.
What data should staff never paste into AI tools?
Passwords, keys and security settings should never go into any AI tool. Personal and sensitive information, such as health records, tax file numbers and payslips, should never go into free or personal accounts, and only into business tools for use cases the privacy lead has approved. Pricing, tenders and board papers belong only in tools approved for that classification.
Does an AI use policy need to cover automated decisions under the Privacy Act?
It should. From 10 December 2026, APP 1.7 to 1.9 require an organisation's privacy policy to explain certain decisions made or substantially assisted by computer programs using personal information. An AI use policy that requires every AI use affecting individuals to be registered gives the privacy lead the list needed to update the privacy policy accurately.
How often should an AI use policy be reviewed?
Review it whenever the approved tool list, a vendor's data terms or the law changes, and at least once a year otherwise. AI products change plans and features often, so the approved tool list and data class examples usually need updating more often than the core principles. Record each review date and owner in the policy itself.
Have this problem across your branches?
A 30-minute review with a senior consultant. You leave with a written one-page summary you can share internally.
Fixed-fee phases, agreed before we start. Stop after any phase.