AI Mistakes Australian Public Servants Should Avoid

AI tools offer real benefits for public servants, but using them without proper care creates risks — for the public servant, for their agency, and for the citizens they serve. These are the most common mistakes Australian public servants make with AI — and how to avoid them.

---

Mistake 1: Using Unapproved AI Tools for Work

Many public servants have started using AI tools — ChatGPT, Claude, Gemini — for work tasks without checking whether their agency has approved these tools. Using unapproved tools creates risks around data security, privacy, and accountability.

The free versions of consumer AI tools are not appropriate for government work. They may process data on servers outside Australia, use prompts to train the underlying model, and lack the security certifications required for government use.

How to avoid it: Check what AI tools your agency has approved before using any AI tool for work. If your agency has not yet published guidance, ask your IT team or manager. Use only approved tools for work purposes.

---

Mistake 2: Entering Sensitive or Classified Information into AI Tools

AI tools process the information you enter into them. Entering classified, sensitive, or personal information into an AI tool that has not been approved for those purposes is a serious security and privacy risk.

This includes: cabinet-in-confidence material, personal information about citizens, commercially sensitive information, information about security operations, and draft policy that has not been cleared for wider circulation.

How to avoid it: Before entering any information into an AI tool, ask yourself: what classification is this information? Has this tool been approved for information at this classification? If in doubt, do not enter it.

---

Mistake 3: Publishing AI-Generated Content Without Review

AI tools can produce plausible-sounding but inaccurate content. A public servant who publishes AI-generated information — on a government website, in a ministerial brief, in a response to a parliamentary question — without reviewing it for accuracy is creating significant risk.

Government information must be accurate. Inaccurate government information can mislead citizens, create legal liability, and damage the agency's reputation.

How to avoid it: Treat AI output as a first draft, not a finished product. Every piece of AI-generated content that will be used in an official document or communication must be reviewed by a staff member with knowledge of the subject matter.

---

Mistake 4: Delegating Decisions to AI

Public servants are accountable for the decisions they make. Using AI to make decisions — about citizen eligibility, compliance assessments, enforcement actions — without human review is not appropriate and may be unlawful.

The Administrative Decisions (Judicial Review) Act 1977 and related legislation require that administrative decisions be made by authorised decision-makers, not automated systems. AI can assist with the analysis and recommendation, but the decision must be made by a qualified officer.

How to avoid it: Be clear about the distinction between AI-assisted analysis and AI-made decisions. AI provides information and drafts; humans make decisions and are accountable for them.

---

Mistake 5: Failing to Disclose AI Use Where Required

Some agencies and some document types require disclosure of AI use. Failing to disclose AI use where it is required is a breach of agency policy and, in some contexts, may raise accountability concerns.

How to avoid it: Check your agency's policy on AI disclosure. As a general principle, if AI was used to draft a significant portion of an official document, this should be noted. Be transparent about how AI tools are being used in your work.

---

Mistake 6: Over-Relying on AI for Complex Policy Analysis

AI tools are good at producing well-structured, plausible-sounding text. They are not good at the kind of deep analytical thinking that complex policy problems require — weighing competing values, understanding political context, anticipating second-order effects.

A public servant who relies on AI to do their policy thinking — rather than using AI to assist with the mechanical aspects of research and writing — is likely to produce lower-quality analysis.

How to avoid it: Use AI for the mechanical aspects of policy work — summarising documents, drafting sections, organising information. Do the analytical thinking yourself. The recommendation and the reasoning behind it must be yours.

---

Mistake 7: Ignoring Bias in AI Output

AI tools can reflect biases present in their training data. In a government context, this could mean AI-generated policy analysis that reflects particular ideological assumptions, or AI-generated citizen communications that are not inclusive.

How to avoid it: Review AI output critically, particularly for policy documents and citizen communications. Be aware that AI tools are not neutral — they reflect the data they were trained on.

---

Mistake 8: Not Keeping Records of AI Use

Public servants are subject to records management obligations. If AI is used to produce official documents, correspondence, or reports, there should be a record of this. This is important for accountability and for responding to FOI requests and audit queries.

How to avoid it: Follow your agency's records management policy for AI-generated content. This might include noting in document metadata that AI was used in drafting, or maintaining a log of AI tool use for significant documents.

---

Mistake 9: Using AI for Sensitive Citizen Interactions

Some citizen interactions require a human touch — complex welfare cases, complaints about government decisions, interactions with vulnerable citizens. Using AI to handle these interactions without significant human involvement risks producing responses that are impersonal, inaccurate, or inappropriate.

How to avoid it: Identify the types of citizen interactions where AI assistance is appropriate and where it is not. Complex, sensitive, or high-stakes interactions should be handled primarily by humans.

---

Mistake 10: Assuming AI Tools Are Up to Date

AI tools are trained on data up to a certain date. They may not reflect the most current legislation, policy, or guidance. A public servant who relies on AI for information about current legal requirements without verifying against current sources is taking a risk.

How to avoid it: Always verify legal and policy requirements against current authoritative sources — legislation.gov.au, agency policy documents, and current guidance from central agencies. Do not rely on AI for information about current legal requirements.