In short
AI compliance spans at least 6 areas and starts with a register
- AI compliance in Australia spans at least 6 areas, and a gap in any one creates exposure. They are privacy, consumer protection, anti-discrimination, workplace relations, industry rules and the disclosure rules in Australian Privacy Principle (APP) 1.7 and APP 1.8.
- Applied AI Australia's AI Roadmap covers governance foundations, and its APP 1.7 Rapid Review covers the automated-decision disclosure gap.
- An AI register that records every deployment, data source and compliance assessment is the first compliance step to take.
- Building compliance into AI deployment from the start costs much less than fixing a non-compliant system after a regulator asks questions.
Detail
Compliance checks for each area of law AI touches
Many organisations treat AI compliance as a single requirement. In Australia it touches at least 6 areas of law, and missing any one of them creates exposure.
Privacy and data protection: for organisations the Privacy Act 1988 covers, collection, use and disclosure of personal information through AI meet the Act. Cross-border disclosures are managed under APP 8. Requests to access or correct personal information are handled even when AI processed the data. Privacy impact assessments are completed for high-risk AI, to identify how each system could affect people's privacy. The Office of the Australian Information Commissioner (OAIC) encourages organisations to complete them. Under the Australian Government Agencies Privacy Code, agencies covered by the Privacy Act must complete one for every high privacy risk project.
Consumer protection: AI-generated advertising and representations are accurate and do not mislead. Pricing algorithms comply with the Australian Consumer Law. Customer service AI does not mislead customers about whether they are dealing with a person. Complaint handling works for decisions made with AI.
Anti-discrimination: hiring AI is tested for bias, and credit and insurance AI is assessed for fairness. Customer-facing AI does not discriminate on protected grounds. Liability can arise even when discrimination was unintended.
Workplace relations: employee monitoring is checked against state and territory workplace surveillance laws and disclosed to staff. AI used in employment decisions is documented and reviewable. Workplace safety duties are met where AI controls physical systems.
Industry rules: Australian Securities and Investments Commission (ASIC) guidance applies to financial services. Australian Prudential Regulation Authority (APRA) standards apply to the entities it regulates. Therapeutic Goods Administration (TGA) requirements apply where AI software is a medical device.
Automated-decision disclosure: from 10 December 2026, APP 1.7 and APP 1.8 set privacy policy disclosure rules for some automated decisions. Our article on what APP 1.7 requires sets out the test in full.
Two Applied AI Australia offers cover part of this work. The AI Roadmap builds governance foundations: frameworks, an AI register, board reporting and the order of work. The APP 1.7 Rapid Review maps the decisions that may be in scope, for your privacy and legal teams. Privacy impact assessments, vendor contract reviews and bias audits are work for your privacy, legal and specialist advisers.
The first step is to build and maintain an AI register. It should record every AI system, data source, purpose, compliance assessment and risk rating. If a regulator asks what AI you run and you cannot answer fully, the rest of your compliance work is hard to show.
Commercial impact
Regulators apply existing law to AI from several directions
Regulators are applying existing law to AI from several directions. The OAIC has published its regulatory action priorities for 2025-26. They name practices that erode privacy rights in the application of artificial intelligence. Consumer law applies to algorithmic pricing, and anti-discrimination law applies to AI in hiring. APP 1.7 to 1.9 add a fixed date for privacy policy disclosure.
Showing compliance across all 6 areas needs a register, assessments and reporting.
The costs are uneven. A privacy impact assessment before deployment costs much less than fixing problems afterwards. Fixing a Privacy Act breach after the OAIC has acted costs more again, before any damage to reputation. Building compliance into AI deployment from the start is cheaper than fixing it later.
For boards, compliance is also a director's concern. Section 180 of the Corporations Act 2001 sets the duty of directors and officers to act with care and diligence. ASIC's Report 798 (October 2024) states that directors' and officers' duties of reasonable care and diligence 'extend to the adoption, deployment and use of AI'. A record that the board asked the right questions and received reporting on AI compliance is evidence of that oversight.
Compliance gaps are one reason AI pilots fail to reach production. When privacy, bias or transparency issues surface late, the organisation either fixes the system at high cost or abandons the project. Dealing with compliance early avoids both outcomes.
In practice
Illustrative cases of AI registers and compliance assessments
Illustrative composite: a large company built an AI compliance framework with its privacy, legal and specialist advisers. It created an AI register covering every deployment and set up quarterly compliance reviews. It required privacy impact assessments for any AI processing customer data, and bias testing for hiring and credit AI. It also introduced vendor compliance audits and board reporting templates.
Illustrative composite: a mid-market retailer deployed AI dynamic pricing without a compliance assessment. The Australian Competition and Consumer Commission (ACCC) raised concerns about misleading price representations created by the algorithm. The retailer spent months fixing the problem.
Illustrative composite: a professional services firm built an AI register and had its privacy and legal advisers complete privacy impact assessments for its high-risk systems. The advisers also reviewed its vendor contracts. Some AI vendors had terms allowing them to use client data to train their models. The firm renegotiated those terms before any client data was used that way.
Illustrative composite: a financial services organisation mapped its customer-facing AI systems before 10 December 2026. It listed the decisions those systems influence about customers. Some needed only a clear description in the privacy policy. Others needed more work: nobody could yet describe how the credit decisioning system reached its decisions, and the fraud system's role had never been documented. Starting early gave its privacy and legal teams time to act.
What to do
Steps to build compliance in before AI goes live
- Build and maintain an AI register as your first compliance step, covering every system, data source, purpose and risk rating.
- Map every AI deployment against all 6 areas of law before it goes live, including privacy, consumer and anti-discrimination law.
- Match the approach to your maturity: the AI Roadmap for governance foundations, and the APP 1.7 Rapid Review for the automated-decision gap.
- Set a date for reviewing your privacy policy against APP 1.7 and APP 1.8, well before 10 December 2026.
- Ask for a compliance assessment in every AI business case, before money is committed.
Last reviewed
Put this topic against your own numbers, owners and timelines.