Victorian administrative guideline on the safe and responsible use of generative AI

What it is

The Administrative Guideline for the safe and responsible use of generative AI in the Victorian Public Sector advises the Victorian Public Sector (VPS) on using generative AI tools for official work. It takes a principles-based approach, given how fast the technology is changing, and sets minimum requirements. The Victorian Secretaries Board endorsed it on 24 September 2024, and the Secretary of the Department of Premier and Cabinet made it on 27 November 2024.

The guideline defines generative AI as AI that creates new content, such as text, images, code and audio, in response to prompts. It names tools including ChatGPT, Google Gemini, Midjourney, Copilot (including Microsoft 365, Edge and GitHub), DALL-E, Synthesia, Perplexity and QuillBot.

Who it applies to

All public service bodies and public entities as defined in the Public Administration Act 2004. It also applies to every employee, contractor, consultant and volunteer engaged directly or indirectly by those organisations who has access to public sector information. Special bodies are encouraged to adopt it where no organisation-specific or sector-specific policy exists. It is read alongside any stricter organisational or sector policies. Schools, for example, apply a Department of Education guideline.

What it requires

The guideline lets in-scope organisations and their personnel use generative AI for official work subject to these safeguards.

• Agency-approved generative AI tools are used ahead of publicly available ones.
• Only publicly available information can be shared with, input into or uploaded to a tool the organisation has not approved.
• Information entered into an approved tool must not exceed the protective marking the organisation has set for that tool.
• Organisations make sure personnel understand their responsibilities and the risks of approved and public tools.
• Organisations monitor the use of approved and public tools to support safe and responsible use.
• Personnel remain responsible and accountable for the accuracy and quality of their work, including advice, decisions and content.
• Personnel continue to meet all their obligations, including under the Privacy and Data Protection Act 2014, the Victorian Protective Data Security Standards, the Health Records Act 2001, the Freedom of Information Act 1982, the Charter of Human Rights and Responsibilities Act 2006, the Public Records Act 1973 and the public sector codes of conduct.

The guideline also says generative AI tools should not be used to make decisions, undertake assessments or carry out other administrative actions. Everyone in scope must consider Australia’s AI Ethics Principles before using generative AI for official work.

Implementation and consequences

Organisations educate their personnel about the guideline. They may monitor use in line with the Office of the Victorian Information Commissioner’s guiding principles for surveillance. Suggested approaches include just-in-time notices when staff open a generative AI tool on the organisation’s network, reporting drawn from network monitoring and training on benefits, risks and responsibilities. The guideline says the Department of Government Services will develop training modules that organisations can adapt.

Failing to meet the guideline may breach the codes of conduct and legislation. Relying on generative AI content to support a decision without checking its accuracy could breach the code of conduct for VPS employees. An organisation that intends to operate inconsistently with the guideline must give written reasons to the Secretary of the Department of Premier and Cabinet under section 36A(3) of the Public Administration Act 2004.

Key dates

Victoria agreed to the national framework for the assurance of AI in government on 21 June 2024. The guideline was endorsed on 24 September 2024 and made on 27 November 2024. Its web page was last updated on 12 February 2025. The guideline will be reviewed periodically, with changes returned to the Victorian Secretaries Board.

How it relates to other frameworks

The guideline is how Victoria applies the national framework for the assurance of AI in government and Australia’s AI Ethics Principles to generative AI. It is narrower than the NSW AI Assessment Framework or the Queensland FAIRA, which require risk assessment of AI use cases in general. Its rules on approved tools and monitoring are close to the acceptable use policies many private organisations write, and to the OAIC’s advice not to put personal information into public generative AI tools.