TEMPLATE: AI POLICYAI policy template for Australian businesses: set the rules before staff set their own

1. Purpose

This policy sets out how people at [Organisation] may use artificial intelligence (AI) at work: which tools, with what data, and who checks the results. It exists so that staff can use AI to do better work while we protect our customers, our people, our information and our reputation.

It follows the Australian Government’s Guidance for AI Adoption and our obligations under the Privacy Act 1988 (Cth). [If you are regulated by APRA: it also supports our obligations under Prudential Standards CPS 234 Information Security and CPS 230 Operational Risk Management.]

2. Scope

• Who: all employees, contractors, consultants and temporary staff, and anyone else using [Organisation] systems or information ("staff").
• What: AI assistants and chatbots (for example ChatGPT, Claude, Gemini and Microsoft Copilot), AI features inside other software (for example meeting transcription, email drafting and CRM assistants), browser extensions and plug-ins that use AI, AI agents that can take actions in other systems, and AI models we build or run ourselves.
• Where: on company devices, and on personal devices whenever [Organisation] information is involved.
• Out of scope: [for example spam filters and other long-established automated tools, listed as out of scope in the AI register].

3. Principles

• People own the outcome. AI output is a draft. The person who uses it is accountable for it.
• Protect data first. Only put information into an AI tool that the tool is approved to receive.
• Be open. Say when AI has materially shaped work that others rely on, and never present AI as a person.
• Use approved tools. If the tool you need is not approved, ask. Do not work around this policy.
• Report early. Mistakes and near misses are reported quickly and handled without blame.

4. Roles and responsibilities

5. Approved tools

• The AI register lists every approved AI tool, its owner, what it may be used for and the highest data class it may receive. [Link to the register.]
• Use approved tools only through your [Organisation] account, with single sign-on where available. Do not use personal AI accounts for work.
• Free and consumer versions of AI tools are approved for Public information only, unless the register says otherwise.
• Requesting a new tool: submit [form or ticket] to [IT]. We aim to answer within [10] business days. The assessment covers where data is stored and processed, whether the vendor uses inputs to train its models, security certifications, logging, contract terms, and how we would exit.
• AI agents that can send email, change records, make payments or act in other systems need approval from the AI governance owner before use. Each agent has a named owner, its own identity with the least access it needs, a log of its actions, and a way for a person to stop it.

6. Information classes and what may go into AI tools

Never put the following into an AI tool that is not approved for it:

• personal information about customers, staff or anyone else, and especially sensitive information such as health information;
• passwords, API keys, access tokens or any other credentials;
• payment card, bank account or tax file numbers;
• information covered by a confidentiality agreement or legal privilege;
• unreleased financial results, deal information or other market-sensitive information;
• source code, security settings or network details, unless the tool is approved for them;
• [anything else specific to your organisation].

If you are unsure of the class, treat the information as the higher class and ask [contact].

7. Human review

• Treat AI output as a draft. Check facts, figures, names, quotes, citations and calculations against the source before you use them.
• You are accountable for anything you send, publish or decide, whether or not AI helped.
• High-risk uses need a named person to make the decision. These include decisions that significantly affect a person (for example credit, employment, claims, pricing for an individual or access to a service), legal or regulatory advice, safety-related work, and anything sent to customers without review. AI may assist; a person decides.
• Do not use AI to make or substantially help make decisions about individuals unless the use is approved, recorded in the AI register and described in our privacy policy (see section 8).
• Label AI-generated content where people could otherwise be misled, and never let a chatbot present itself as a person.

8. Privacy

• The Privacy Act 1988 and the Australian Privacy Principles (APPs) apply to personal information we put into an AI tool, and to AI output that contains personal information.
• Following guidance from the Office of the Australian Information Commissioner (OAIC), do not enter personal information, and particularly sensitive information, into publicly available generative AI tools.
• Use and disclosure (APP 6): use personal information in an AI tool only for the purpose it was collected for, for a related purpose the person would reasonably expect, or with consent.
• Generated and inferred information (APP 3): if AI generates or infers information about an identifiable person, that is a collection of personal information and must be reasonably necessary and done by lawful and fair means.
• Accuracy and security (APP 10 and APP 11): check AI output about people before relying on it, and use only tools that meet our security requirements.
• Overseas processing (APP 8): many AI tools process data outside Australia. Before personal information goes to an AI tool, the privacy officer confirms where it is processed and that the disclosure is allowed.
• Automated decisions (APP 1): from 10 December 2026, our privacy policy must describe the kinds of personal information used, and the kinds of decisions made, where a computer program makes, or does something substantially and directly related to making, a decision that could significantly affect a person. Tell the privacy officer before starting any AI use of this kind.
• A privacy impact assessment is required before any new AI use that involves personal information.

9. Confidentiality and intellectual property

• Approved tools must not use our inputs to train the vendor’s models unless the AI governance owner has agreed in writing.
• Do not upload material owned by others, including client documents, beyond what our contract or licence allows.
• AI output can resemble existing work, and ownership of AI-generated material can be uncertain. Ask [legal] before relying on AI output for anything we need to own, license or publish under our name.

10. Records

• Keep AI prompts and outputs when they form part of a business record, for example advice to a customer, a decision about a person or a regulatory submission. Store them in [system] for the period set by our records policy.
• Approved tools keep logs of who used them and when, and, where practical, the prompts and outputs. Logs are kept for [period].
• For each AI tool or system, the AI register records: owner, purpose, highest data class, vendor, model and version where known, where data is processed, risk rating and last review date.

11. Incidents

Report to [contact] straight away, and within [24 hours] at most, if:

• information went into an AI tool that was not approved to receive it;
• AI output caused, or nearly caused, harm to a customer, a colleague or [Organisation];
• an AI tool or agent did something unexpected, such as reading data or taking an action it should not have;
• you suspect a manipulated prompt (prompt injection), a compromised AI account or a deepfake.

[Security] contains the incident and records it. The privacy officer assesses whether it is an eligible data breach under the Notifiable Data Breaches scheme, which requires us to notify affected individuals and the OAIC when a breach is likely to result in serious harm. [If you are regulated by APRA: material information security incidents must be notified to APRA within 72 hours under CPS 234.] Prompt, honest reporting is never penalised.

12. Training

• Everyone completes AI training before using approved AI tools, and refreshes it [every 12 months].
• Training covers this policy, the information classes, how to check AI output, prompt injection, AI-assisted scams and deepfakes, and how to report incidents.
• People who oversee higher-risk AI uses receive role-specific training. [The board and executive receive an AI briefing at least once a year.]

13. Breaches of this policy

Breaches are handled under [code of conduct or disciplinary policy]. Honest mistakes that are reported promptly are treated as a chance to improve our controls.

14. Review

• The AI governance owner reviews this policy at least every [6 months], and also after a significant AI incident, a new type of AI use (such as AI agents), a change in law or regulator guidance, or a major vendor change.
• Changes are recorded in the version history below and communicated to all staff.

15. Acknowledgement

I have read and understood the [Organisation] Acceptable Use of AI Policy and agree to follow it.

Name: ____________________ Signature: ____________________ Date: ____________