In short
High-impact AI that nobody can explain is hard to defend
- From 10 December 2026, Australian Privacy Principle (APP) 1.7 and APP 1.8 set privacy policy disclosure rules for some automated decisions.
- If your AI makes decisions that affect people, you need to be able to describe them. Examples include credit, hiring, pricing and claims decisions. Explaining individual outcomes also reduces legal, complaint and trust risk.
- High-impact AI that nobody can explain is hard to defend. An answer such as 'the algorithm decided' is unlikely to satisfy customers, complaint bodies or regulators.
- Transparency also builds trust. Customers, employees and regulators are more willing to accept AI decisions they can understand.
Detail
Transparency covers systems, decision logic and accountability
From 10 December 2026, APP 1.7 and APP 1.8 set privacy policy disclosure rules for some automated decisions. They apply to APP entities, which are organisations and agencies covered by the Privacy Act 1988. Our article on what APP 1.7 requires sets out the test in full.
To describe your decisions, you need to understand how they work. That means plain language a customer, employee or regulator can follow, as well as technical detail.
The date is fixed. Organisations that start now have time to find, understand and describe their decisions.
Transparency covers 3 layers. System documentation records what AI systems exist, what they do, what data they use and what decisions they inform or automate. Decision logic explains how the AI reaches conclusions, what factors it weighs and why it recommended an outcome for a particular person. Human accountability names who owns the AI's decisions, who can override them and how affected people can seek review.
Other laws already apply. APP 1 requires open and transparent management of personal information. Consumer law prohibits misleading and unconscionable conduct. Anti-discrimination law applies to hiring and other decisions about people. APP 1.7 to 1.9 add an explicit disclosure duty on top.
Not all AI needs the same level of explanation. A chatbot suggesting products needs basic disclosure. AI that decides credit applications should give the factors behind each decision and a path to a person. AI that screens job applicants should have documented logic, bias testing and a way to challenge the result. The principle is proportionality: the greater the impact on a person, the more transparency is needed.
For boards, the test is whether the organisation can explain every AI decision that affects a customer, employee or other stakeholder. If it cannot do so for a high-impact system, there is work to do before December.
The Office of the Australian Information Commissioner (OAIC) published updated guidance on APP 1.7 to 1.9 on 30 September 2026. Its regulatory action priorities for 2025-26 also name practices that erode privacy rights in the application of artificial intelligence.
Commercial impact
Unexplained AI creates legal, compliance and trust risk
AI that nobody can explain creates 3 kinds of risk. Legal risk arises because automated decisions can be challenged through complaints, disputes and the courts. If you cannot explain why your AI rejected a credit application or did not shortlist a candidate, you are exposed. Compliance risk arises because the OAIC has named AI in its regulatory action priorities for 2025-26. APP 1.7 to 1.9 also add an explicit disclosure duty. Trust risk arises because people do not trust decisions they cannot understand, and distrust hurts adoption and retention.
Undisclosed AI use inside the organisation becomes a liability once disclosure is a legal requirement. Your privacy policy can describe only the systems people will admit exist.
Boards also need to understand the systems they oversee. If an AI vendor cannot explain how its model reaches decisions, accountability has passed to a system nobody can inspect. The Australian Securities and Investments Commission (ASIC) published Report 798 in October 2024. It states that directors' and officers' duties of reasonable care and diligence 'extend to the adoption, deployment and use of AI'. A vendor's assurance is a weak basis for that oversight.
Transparency also has a competitive side. Organisations that build it into AI from the start meet compliance requirements, build customer trust and improve employee adoption. Their systems are also easier to debug and improve. When you cannot explain why AI reached a conclusion, you cannot validate, correct or improve it.
In practice
Illustrative cases of explainability built in or missing
Illustrative composite: a financial services company built explainability into its credit decision AI from the start. Every decision recorded the main factors behind it, such as the ratio of income to debt, payment history and employment stability. Customer service staff could explain a rejection in plain language. Internal reviews could check the system for discrimination.
Illustrative composite: an insurer's claims AI made decisions that frontline staff could not explain, so staff could tell customers only that the system had decided. Complaints rose, and some disputes reached the Australian Financial Complaints Authority. The insurer then weighed going back to decision rules that staff could explain. That change would cost some accuracy.
Illustrative composite: a retailer's AI pricing system had no explainability at all. When prices moved without visible logic, customers complained and the Australian Competition and Consumer Commission (ACCC) made inquiries. The retailer then introduced decision logging and price explanations for customers.
Illustrative composite: a professional services firm requires a written rationale for every AI-assisted hiring decision. The rationale records the criteria the AI weighted, why candidates were surfaced and what influenced the rankings. Hiring managers are trained to explain it. If a rejected candidate complains of discrimination, the firm can show how the decision was made against its criteria.
What to do
Steps to build explainability into AI that affects people
- Review every AI system against APP 1.7 to 1.9 before 10 December 2026, and plan the work needed now.
- Set explainability requirements for all AI that affects individuals, including credit, hiring, claims, pricing and customer decisions.
- Build transparency into AI from the start, because adding it later can cost more.
- Make sure AI vendor contracts include explainability requirements as well as performance metrics.
- Train customer-facing and HR teams to explain AI decisions in plain language, with clear paths for escalation.
Last reviewed
Put this topic against your own numbers, owners and timelines.