How to use this sample
This is a starting point for a business of any size from a sole trader up to about fifty people. Replace everything in [square brackets], delete what does not apply and keep it short enough that staff will read it.
Each clause notes which of the six practices in the Australian Government’s Guidance for AI Adoption and which of the 10 guardrails in the earlier Voluntary AI Safety Standard it supports. Both are voluntary. This sample is not legal advice, and existing laws such as the Privacy Act 1988 (Cth) still apply.
1. Accountability
1.1 [Name, role] is the AI owner for [Business name]. The AI owner approves new AI tools and uses, keeps the AI-use register, reviews non-conformance reports and reviews this policy.
1.2 Each approved use in the register has a named person who is responsible for how it is used day to day.
1.3 Using an AI tool does not move responsibility for the work. The person who sends, files or acts on AI-assisted work is responsible for it, as if they had written it themselves.
Maps to
- Guidance for AI Adoption, practice 1: Decide who is accountable
- Voluntary AI Safety Standard, guardrail 1
2. Approved tools
2.1 Staff may use only the AI tools and uses listed in the AI-use register for [Business name] work. This includes AI features switched on inside software we already use.
2.2 To add a tool or a new use, complete the new-use approval checklist and send it to the AI owner. Do not start using it until the AI owner has approved it in writing. This is a hold point.
2.3 Before approving, the AI owner checks: what information the tool will receive, where the provider stores it, whether it is used to train the provider’s models, whether a business account is available, and what the tool’s terms say about confidentiality.
2.4 Use business accounts set up by [Business name], not personal accounts, for anything above green information (see clause 3).
Maps to
- Guidance for AI Adoption, practice 1: Decide who is accountable
- Guidance for AI Adoption, practice 3: Measure and manage risks
- Voluntary AI Safety Standard, guardrail 2
- Voluntary AI Safety Standard, guardrail 3
- Voluntary AI Safety Standard, guardrail 8
3. Information you may enter
3.1 Before you enter anything into an AI tool, classify it:
| Class | Examples | Rule |
|---|---|---|
| Green | Published information, general questions, your own de-identified drafts | Any approved tool |
| Amber | Internal procedures, pricing, supplier details, non-public business plans | Approved business-account tools only |
| Red | Customer or staff personal information, health information, financial account details, anything under a confidentiality agreement, passwords and keys | Not permitted unless the AI-use register lists that exact use, tool and safeguard |
3.2 If you are unsure, treat it as red and ask the AI owner.
3.3 The Office of the Australian Information Commissioner’s guidance is that, as a matter of best practice, personal information, and especially sensitive information, should not be entered into publicly available generative AI tools.
Maps to
- Guidance for AI Adoption, practice 3: Measure and manage risks
- Voluntary AI Safety Standard, guardrail 3
4. Checking AI output
4.1 Treat AI output as a draft from a fast but unreliable junior. Check facts, figures, names, dates, references and quotes against the source before relying on them.
4.2 A named person must check every AI-assisted output before it is sent to a customer, supplier or regulator, or used in a decision about a person.
4.3 Where the register sets acceptance criteria for a use (for example “every price matches the current price list”), check against those criteria every time.
4.4 Anyone may stop using an AI tool for a task if they think its output is unsafe, and should tell the AI owner.
Maps to
- Guidance for AI Adoption, practice 5: Test and monitor
- Guidance for AI Adoption, practice 6: Maintain human control
- Voluntary AI Safety Standard, guardrail 4
- Voluntary AI Safety Standard, guardrail 5
5. Telling people
5.1 Tell customers when they are dealing with an AI system rather than a person (for example a chat assistant on our website).
5.2 Do not present AI-generated images, audio or video of real people or places as real.
5.3 If a customer asks whether AI was used in work we did for them, answer honestly.
Maps to
- Guidance for AI Adoption, practice 4: Share essential information
- Voluntary AI Safety Standard, guardrail 6
6. When something goes wrong
6.1 If AI output was wrong and was used, or information was entered in breach of clause 3, record a non-conformance report (NCR) in the NCR log the same day. Recording an NCR is expected and is not a disciplinary matter.
6.2 Each NCR records what happened, who was affected, the immediate fix (for example, correcting and resending), the cause and the change made to stop it recurring (a prompt, a check, a rule or training).
6.3 If personal information may have been disclosed, tell the AI owner immediately. They will assess whether it is a notifiable data breach and take advice where needed.
6.4 If a customer questions an outcome that AI contributed to, a person reviews it and responds.
Maps to
- Guidance for AI Adoption, practice 3: Measure and manage risks
- Guidance for AI Adoption, practice 5: Test and monitor
- Voluntary AI Safety Standard, guardrail 2
- Voluntary AI Safety Standard, guardrail 7
7. Not permitted
7.1 Do not use AI to:
- make a final decision about hiring, discipline, pay, credit or service eligibility without a person making that decision;
- impersonate a real person, or write fake reviews;
- produce advice we are not qualified to give (for example legal or medical advice);
- get around a security control or another policy.
Maps to
- Guidance for AI Adoption, practice 2: Understand impacts and plan accordingly
- Guidance for AI Adoption, practice 6: Maintain human control
- Voluntary AI Safety Standard, guardrail 5
- Voluntary AI Safety Standard, guardrail 10
8. Training, records and review
8.1 Everyone who uses AI tools for [Business name] reads this policy and attends a short briefing before starting, and when it changes.
8.2 The AI owner keeps the AI-use register, approvals and NCR log for [period set by your record-keeping policy].
8.3 The AI owner reviews this policy every [six] months, and whenever a tool, use or law changes, and asks the people affected by AI-assisted work (staff and, where relevant, customers) whether it is working for them.
Approved by: [Name, role] · Date: [date] · Next review: [date]
Maps to
- Guidance for AI Adoption, practice 1: Decide who is accountable
- Guidance for AI Adoption, practice 2: Understand impacts and plan accordingly
- Guidance for AI Adoption, practice 4: Share essential information
- Voluntary AI Safety Standard, guardrail 1
- Voluntary AI Safety Standard, guardrail 9
- Voluntary AI Safety Standard, guardrail 10
Mapping summary
| Clause | Guidance for AI Adoption practices | VAISS guardrails |
|---|---|---|
| 1. Accountability | 1 | 1 |
| 2. Approved tools | 1, 3 | 2, 3, 8 |
| 3. Information you may enter | 3 | 3 |
| 4. Checking AI output | 5, 6 | 4, 5 |
| 5. Telling people | 4 | 6 |
| 6. When something goes wrong | 3, 5 | 2, 7 |
| 7. Not permitted | 2, 6 | 5, 10 |
| 8. Training, records and review | 1, 2, 4 | 1, 9, 10 |
The six practices
- Decide who is accountable
- Understand impacts and plan accordingly
- Measure and manage risks
- Share essential information
- Test and monitor
- Maintain human control
The 10 guardrails
- Establish, implement and publish an accountability process including governance, internal capability and a strategy for regulatory compliance.
- Establish and implement a risk management process to identify and mitigate risks.
- Protect AI systems, and implement data governance measures to manage data quality and provenance.
- Test AI models and systems to evaluate model performance and monitor the system once deployed.
- Enable human control or intervention in an AI system to achieve meaningful human oversight across the life cycle.
- Inform end-users regarding AI-enabled decisions, interactions with AI and AI-generated content.
- Establish processes for people impacted by AI systems to challenge use or outcomes.
- Be transparent with other organisations across the AI supply chain about data, models and systems to help them effectively address risks.
- Keep and maintain records to allow third parties to assess compliance with guardrails.
- Engage your stakeholders and evaluate their needs and circumstances, with a focus on safety, diversity, inclusion and fairness.
Sources
- Guidance for AI Adoption (opens in a new tab)
The current Australian Government guidance. It updates the Voluntary AI Safety Standard and condenses its 10 guardrails into six essential practices. Voluntary.
- Guidance for AI Adoption: Foundations (opens in a new tab)
The short version for organisations starting out or using AI in lower-risk ways. This is the version most SMEs should read first.
- Voluntary AI Safety Standard: the 10 guardrails (opens in a new tab)
The guardrail wording used in the mapping tables on this site.
- Guidance on privacy and the use of commercially available AI products (opens in a new tab)
Explains how the Privacy Act 1988 (Cth) applies when a business puts personal information into an AI product. Its best-practice position is not to enter personal information, especially sensitive information, into publicly available generative AI tools.
Free to copy and adapt. The AI Use-Case Assessment produces the AI-use register this policy refers to, and the pilot produces the release records.