AI Privacy Lawyers Adelaide: Privacy Act Rules for AI Tools, Automated Decisions and Training Data
AI and privacy in South Australia
For businesses covered by the Privacy Act, it applies to personal information going into an AI tool and coming out of it. From 10 December 2026, privacy policies must also explain automated decisions. Talon Legal advises Adelaide businesses that use AI tools on customer information, and businesses building AI into their products. Book a same-day free 30-minute consultation or call (08) 7094 2021.
AI and privacy key facts
| Issue | Position in South Australia |
|---|---|
| Who is covered | APP entities. A business with annual turnover of $3,000,000 or less is generally exempt unless an exception applies (Privacy Act 1988 (Cth) section 6D). |
| OAIC guidance on AI products | The OAIC's Guidance on privacy and the use of commercially available AI products (21 October 2024, updated 17 January 2025) says the Privacy Act applies to all uses of AI involving personal information, and that public-facing chatbots should be clearly identified as such. |
| AI-generated content | Personal information includes an opinion, whether true or not, about an identified or reasonably identifiable individual (section 6(1)). |
| Prompts and uploads | A use or disclosure for a secondary purpose needs consent or an exception, such as reasonable expectation and relatedness (APP 6.1 and 6.2(a)). |
| Automated decisions | From 10 December 2026, a privacy policy must describe decisions, made by computer programs using personal information, that could reasonably be expected to significantly affect individuals (APP 1.7 and 1.8). |
| Training on data held | Where consent or reasonable expectation cannot be clearly established, the OAIC's Guidance on privacy and developing and training generative AI models (21 October 2024, updated 23 October 2024) says seek consent, offer a meaningful and informed opt-out, or both. |
| Overseas AI vendors | Reasonable steps before disclosing to an overseas recipient (APP 8.1). Where the APPs do not otherwise apply to the recipient, its act that would breach them is taken to be the business's (section 16C). |
| Maximum penalty | Serious interference with privacy by a body corporate: the greatest of $50,000,000, three times the benefit if it can be determined, or otherwise 30% of adjusted turnover during the breach turnover period. Anyone else: $2,500,000 (section 13G). |
| First step | Same-day free 30-minute consultation. Call (08) 7094 2021. |
Need an AI privacy lawyer in Adelaide?
Our AI privacy lawyers in Adelaide advise businesses that use AI tools on personal information, and businesses that build AI into their products. The Privacy Act 1988 (Cth) and its Australian Privacy Principles (APPs) govern that information. The Privacy and Other Legislation Amendment Act 2024 adds rules on automated decisions from 10 December 2026. The Office of the Australian Information Commissioner (OAIC) has published Guidance on privacy and the use of commercially available AI products (21 October 2024, updated 17 January 2025) and Guidance on privacy and developing and training generative AI models (21 October 2024, updated 23 October 2024). This page states the law current to 7 November 2026.
This guide is part of our artificial intelligence practice. Privacy Act compliance generally is covered in our data protection and privacy guide. Related guides cover AI contracts and data licensing, AI and copyright and cyber incident response.
Free Consultation. Before staff use a new AI tool on customer information, send us its terms. Book a same-day free 30-minute consultation or call (08) 7094 2021.
On this page
- Does the Privacy Act apply to a business using AI?
- Are AI outputs and inferences personal information?
- Can staff put personal information into an AI tool?
- What must a privacy policy say about AI?
- What changes for automated decisions from 10 December 2026?
- Can a business train AI on personal information it holds?
- What security does an AI system need?
- What if the AI vendor is overseas?
- What are the penalties?
- Can individuals sue for a serious invasion of privacy?
- How similar matters tend to play out
- How we act on AI and privacy
- Case law
- Frequently asked questions
Does the Privacy Act apply to a business using AI?
Yes, if the Act covers the business at all. An APP entity must not breach an APP (section 15). The OAIC’s Guidance on privacy and the use of commercially available AI products (21 October 2024, updated 17 January 2025) says the Privacy Act applies to all uses of AI involving personal information.
Small businesses are generally outside the Act. A business with annual turnover of $3,000,000 or less for the previous financial year is a small business (section 6D(1)), and a small business operator is not an organisation bound by the APPs (section 6C). The exemption is lost by a business that provides a health service and holds health information, other than in an employee record. It is also lost by one that discloses personal information for a benefit, service or advantage, or provides one to collect personal information from anyone else (section 6D(4)(b) to (d)). A small business whose AI product sells user data, or discloses it for another benefit, is covered that way.
Where in doubt, the OAIC’s Guidance on privacy and developing and training generative AI models (21 October 2024, updated 23 October 2024) says to assume the Privacy Act applies. The data protection and privacy guide explains who the Act covers.
Are AI outputs and inferences personal information?
Often, yes. Personal information is information or an opinion about an identified or reasonably identifiable individual (section 6(1)). It counts whether true or not, and whether recorded in a material form or not. An AI output about a named customer can be personal information even when it is wrong.
The OAIC’s Guidance on privacy and the use of commercially available AI products (21 October 2024, updated 17 January 2025) says privacy obligations apply to personal information put into an AI system. They also apply to AI output that contains personal information. The guidance says inferred, incorrect or artificially generated information about an identifiable person is personal information. Its examples are hallucinations and deepfakes. It also says that generating or inferring personal information with AI, including images, is a collection that must comply with APP 3. The guidance says such a collection must be reasonably necessary and done by lawful and fair means. On that view, a tool that profiles customers is collecting personal information.
Under APP 3.2, an organisation must not collect personal information, other than sensitive information, unless it is reasonably necessary for one or more of its functions or activities. Sensitive information also needs consent, unless APP 3.4 applies (APP 3.3). The question is whether the business needs what the tool infers.
Can staff put personal information into an AI tool?
Only within APP 6. Personal information collected for a primary purpose must not be used or disclosed for a secondary purpose unless the person consents or an exception applies (APP 6.1). One exception covers a use or disclosure the person would reasonably expect, if it is related to the primary purpose (APP 6.2(a)). For sensitive information it must be directly related. Another exception covers uses required or authorised by law (APP 6.2(b)).
The OAIC’s Guidance on privacy and the use of commercially available AI products (21 October 2024, updated 17 January 2025) says this rule applies to inputting personal information into an AI system. The guidance adds that setting the use out in a collection notice and the privacy policy can bear on what people reasonably expect. Its best-practice recommendation is not to enter personal information, particularly sensitive information, into publicly available generative AI tools. It also recommends due diligence before adopting a product, covering testing for the intended uses, human oversight, privacy and security risks, and who can access the personal information. The guidance recommends policies and procedures for AI use as well.
The vendor’s terms settle much of this, as the AI contracts and data licensing guide explains. For staff, the practical control is a written AI use policy. It should name the approved tools and say what information may go into them. Outputs should be checked before anyone relies on them.
Client information can be confidential as well as personal. Equity protects confidential information against actual or threatened misuse without consent. The information must be identified with specificity and have the necessary quality of confidence. It must also have been received in circumstances importing an obligation of confidence (Optus Networks Pty Ltd v Telstra Corporation Ltd [2010] FCAFC 21 at [39]). That protection does not depend on the Privacy Act, so it can bind a small business the Act does not cover.
What must a privacy policy say about AI?
The APPs set the contents. The OAIC’s Guidance on privacy and the use of commercially available AI products (21 October 2024, updated 17 January 2025) says businesses should update privacy policies and notifications with clear information about their use of AI. It also says public-facing AI tools, such as chatbots, should be clearly identified as such.
An APP entity must have a clearly expressed and up-to-date APP privacy policy (APP 1.3). The policy must cover the kinds of personal information collected and held, how and why, how to seek access and correction, and how to complain. It must also say whether information is likely to be disclosed overseas and, if practicable, to which countries (APP 1.4). A breach of APP 1.3 or 1.4 carries up to 200 penalty units, and infringement and compliance notices are available (section 13K(1) and (4)).
What changes for automated decisions from 10 December 2026?
From 10 December 2026, privacy policies must explain certain automated decisions. Schedule 1 Part 15 of the Privacy and Other Legislation Amendment Act 2024 inserts APP 1.7 to 1.9 from 10 December 2026.
From 10 December 2026, an APP privacy policy must contain the APP 1.8 information if three things are true (APP 1.7):
- The entity has arranged for one or more computer programs to make a decision, or to do a thing substantially and directly related to making one.
- The decision could reasonably be expected to significantly affect the rights or interests of an individual.
- Personal information about the individual is used in the operation of those computer programs to make the decision or do that thing.
From 10 December 2026, the policy must then set out three things (APP 1.8):
- The kinds of personal information used in the operation of such computer programs.
- The kinds of such decisions made solely by the operation of such computer programs.
- The kinds of such decisions for which a thing substantially and directly related to making the decision is done by such computer programs.
From 10 December 2026, APP 1.9 widens these terms. Making a decision includes refusing or failing to make one, and doing a thing includes refusing or failing to do it. Rights or interests may be affected adversely or beneficially. One example is a decision under an Act or legislative instrument to grant or refuse a benefit. Another is a decision affecting rights under a contract, agreement or arrangement. A third is a decision affecting access to a significant service or support.
From 10 December 2026, the rules apply to decisions made after commencement (item 89). That is so even if the arrangement, the use of the information or its acquisition came earlier, so systems already running are caught from 10 December 2026. APP 1.7 is also added, from 10 December 2026, to the APPs whose breach can attract an infringement or compliance notice under section 13K (item 87).
To prepare for 10 December 2026, a business using AI to approve, refuse, price or rank people should list the decisions that could significantly affect them, and draft the policy wording.
Can a business train AI on personal information it holds?
Only if APP 6 allows the new use. Suppose training was not a primary purpose of collection. The OAIC’s Guidance on privacy and developing and training generative AI models (21 October 2024, updated 23 October 2024) then says the business needs consent, or reasonable expectation and relatedness, or another exception. Where that cannot be clearly established, it says to seek consent, offer a meaningful and informed opt-out, or both. The guidance makes five further points:
- Publicly available or accessible data is not automatically lawful to use to train or fine-tune a model. Check it for personal information.
- Sensitive information generally requires consent. Many photographs or recordings of people, including artificially generated ones, contain sensitive information.
- Developers must take reasonable steps towards accuracy, such as high quality datasets and testing. Disclaimers may be appropriate.
- APPs 1, 3, 5, 6 and 10 are particularly relevant, and APPs 8, 11, 12 and 13 are also relevant.
- Privacy by design and a privacy impact assessment are recommended.
What security does an AI system need?
Reasonable steps, as for any system holding personal information. An APP entity must take reasonable steps to protect personal information from misuse, interference and loss, and from unauthorised access, modification or disclosure (APP 11.1). Those steps include technical and organisational measures (APP 11.3). An entity must also take reasonable steps to destroy or de-identify information it no longer needs for any permitted purpose. That does not apply to information in a Commonwealth record, or that an Australian law or a court or tribunal order requires it to keep (APP 11.2).
An entity holds personal information if it has possession or control of a record containing it (section 6(1)). Prompt logs and uploaded files can therefore be held information.
The obligation was considered in Australian Information Commissioner v Australian Clinical Labs Limited (No 2) [2025] FCA 1224. Halley J said APP 11.1(b) requires steps reasonable in the circumstances to protect personal information from unauthorised access, modification or disclosure ([49]). The standard is objective, and the circumstances are broadly construed ([50]). They could include the sensitivity of the information, the potential harm and the entity’s size and sophistication. They could also include the cybersecurity environment and previous threats or attacks. Overreliance on third-party providers, without adequate procedures of its own to detect and respond, was one reason the company fell short ([52](g)).
If an AI vendor holding the business’s personal information is breached, the Notifiable Data Breaches scheme can apply to the business (section 26WE). The cyber incident response guide covers the 30-day assessment and notification.
What if the AI vendor is overseas?
APP 8 will usually apply. The OAIC’s APP Guidelines, chapter 8 (updated 3 October 2025), say that engaging an overseas contractor to perform services is, in most circumstances, a disclosure. In limited cases, where a binding contract keeps the business in effective control of the information, providing it may be a use instead.
Before disclosing personal information to an overseas recipient, a business must take reasonable steps so that the recipient does not breach the APPs, other than APP 1 (APP 8.1). APP 8.1 does not apply where the business reasonably believes the recipient is subject to a law or binding scheme that protects the information in a way at least substantially similar to the APPs (APP 8.2(a)). The individual must be able to access mechanisms to enforce that protection.
Where APP 8.1 applied to the disclosure and the APPs do not otherwise apply to the recipient, section 16C applies. An act of the recipient that would breach the APPs (other than APP 1) is taken to be the business’s act, and its breach. Where APP 8.1 applied and the vendor holds the information, Part IIIC applies as if the business held it (section 26WC(1)). The vendor contract should settle data location, use of inputs, subcontractors and breach notice, as the data security clauses guide explains.
What are the penalties?
The Privacy Act has three tiers of civil penalty.
| Contravention | Provision | Maximum |
|---|---|---|
| Serious interference with privacy | Section 13G | Body corporate: the greatest of $50,000,000, three times the benefit if it can be determined, or otherwise 30% of adjusted turnover during the breach turnover period. Anyone else: $2,500,000 |
| Interference with privacy | Section 13H | 2,000 penalty units |
| Breach of APP 1.3 or 1.4, and from 10 December 2026 APP 1.7 | Section 13K | 200 penalty units, with infringement and compliance notices |
These figures are current to 7 November 2026. If a court is not satisfied that an interference was serious, it may order a section 13H penalty (section 13J).
Scale multiplies exposure. In Clinical Labs, the parties agreed, and the Court was satisfied, that there was a separate contravention for each of the more than 223,000 individuals whose information was held on the affected systems ([60]). The Court approved agreed penalties totalling $5,800,000 for three kinds of contravention, $4,200,000 of it for failing to take reasonable steps to protect personal information (orders). An AI system that processes many people’s information carries the same arithmetic.
Can individuals sue for a serious invasion of privacy?
Yes, since 10 June 2025, under Schedule 2 of the Privacy Act. The individual must show an invasion of privacy by intrusion upon seclusion or misuse of information relating to them. There must have been a reasonable expectation of privacy in all the circumstances, and the invasion must have been intentional or reckless, and serious. The public interest in privacy must also outweigh any countervailing public interest (clause 7(1)). Proof of damage is not needed (clause 7(2)).
Damages for non-economic loss and exemplary damages together are capped at the greater of $478,550 and the defamation cap for non-economic loss (clause 11(5)). Proceedings must be started before the earlier of one year after the plaintiff became aware of the invasion and three years after it. A plaintiff who was under 18 at the time must start before their 21st birthday. The court may extend time, but not beyond 6 years after the invasion (clause 14). An AI use that misuses information about a person could found the action if every element is met.
How similar matters tend to play out
Illustrations of how the law on this page combines on ordinary facts, not predictions. Outcome depends on the facts, the evidence and the regulator.
Lower range. A real estate agency with turnover over $3,000,000 finds that property managers paste tenants’ names and arrears into a free public chatbot to draft reminder letters. The OAIC’s Guidance on privacy and the use of commercially available AI products (21 October 2024, updated 17 January 2025) says APP 6 applies to that input. It recommends against entering personal information into publicly available generative AI tools. Whether tenants would reasonably expect it is doubtful (APP 6.2(a)). Moving staff to an approved business tool under a written policy lowers the risk.
Middle range. An online retailer’s website chatbot answers order questions and keeps transcripts. The retailer then uses them to infer customers’ likely income for targeted offers. The OAIC’s Guidance on privacy and the use of commercially available AI products (21 October 2024, updated 17 January 2025) says that inference is a collection. It must then be reasonably necessary for the retailer’s functions or activities (APP 3.2). The guidance also says the chatbot should be identified as AI. The privacy policy must state the purposes for which the retailer collects personal information (APP 1.3 and 1.4).
Upper range. A car subscription business uses an overseas vendor’s AI model, fine-tuned on its records of past subscribers, to approve or refuse applications. From 10 December 2026, if those decisions could reasonably be expected to significantly affect applicants, its privacy policy must describe them (APP 1.7 and 1.8), even though the system started earlier (item 89). The fine-tuning needs consent or another APP 6 basis. If APP 8.1 applied to the disclosure and the APPs do not otherwise apply to the vendor, its breach of the APPs is taken to be the business’s own (section 16C). A serious interference by a body corporate carries the section 13G maximums.
How we act on AI and privacy
Our AI privacy lawyers act for businesses adopting AI tools and for those building them. The aim is a documented basis under the APPs for each AI use, and a privacy policy that meets the rules that apply from 10 December 2026.
| Stage | What happens | What we do |
|---|---|---|
| Mapping | The business lists its AI tools and the personal information each uses or produces. | Identify the APPs that apply to each use, including collection by inference. |
| Vendor review | A tool is chosen or renewed. | Review the terms on use of inputs, storage location, security and breach notice. |
| Policies | Staff start using AI tools. | Draft the AI use policy, and update the privacy policy and collection notices. |
| Automated decisions | Software makes or supports decisions about people. | Advise whether the rules that apply from 10 December 2026 catch them, and draft the policy wording. |
| Training | A model is trained or fine-tuned on data the business holds. | Advise on the APP 6 basis, consent or opt-out, and the privacy impact assessment. |
We cannot promise an outcome, and no lawyer can. We can make sure each AI use has a documented basis under the APPs before the regulator asks for one.
Free Consultation. Book a same-day free 30-minute consultation, or call (08) 7094 2021 now.
AI and privacy case law
No case on this page concerned an AI tool.
| Case | Court | Principle | Pinpoint |
|---|---|---|---|
| Australian Information Commissioner v Australian Clinical Labs Limited (No 2) [2025] FCA 1224 | Federal Court | Reasonable steps to protect personal information; a contravention for each individual | [49], [50], [52](g), [60] |
| Optus Networks Pty Ltd v Telstra Corporation Ltd [2010] FCAFC 21 | Full Federal Court | Equitable obligations of confidence survive a contractual definition | [29], [37], [39] |
Case notes
Australian Information Commissioner v Australian Clinical Labs Limited (No 2) [2025] FCA 1224
The company consented to declarations including that it failed to take reasonable steps to protect personal information on systems it had acquired. Halley J approved agreed penalties of $5,800,000 for three kinds of contravention, $4,200,000 of it for that failure ([7], orders). The steps required are those reasonable in the circumstances, judged objectively ([49], [50]). Overreliance on third-party providers was one reason the company fell short ([52](g)), and there was a separate contravention for each of more than 223,000 individuals ([60]).
Optus Networks Pty Ltd v Telstra Corporation Ltd [2010] FCAFC 21
The Full Court held that an exhaustive contractual definition of confidential information did not show an intention to exclude equitable duties of confidence ([29]). The agreement preserved equitable rights ([37]). The Court set out the four elements of breach of confidence in equity ([39]) and gave judgment for Optus on its claim in equity.
Frequently asked questions
Does the Privacy Act apply to AI tools?
Yes, for businesses the Act covers. An APP entity must not breach an APP (section 15), whatever tools it uses.
Can staff paste customer details into ChatGPT or other AI chatbots?
Only within APP 6. The OAIC’s Guidance on privacy and the use of commercially available AI products (21 October 2024, updated 17 January 2025) recommends not entering personal information, particularly sensitive information, into publicly available generative AI tools.
Is information an AI tool makes up about a person personal information?
Yes, if the person is identified or reasonably identifiable. Personal information includes information or an opinion, whether true or not (section 6(1)).
What must a privacy policy say about automated decisions from 10 December 2026?
From 10 December 2026, it must set out the kinds of personal information used by computer programs that make decisions which could reasonably be expected to significantly affect people. From 10 December 2026, it must also set out the kinds of those decisions (APP 1.7 and 1.8).
Do the rules from 10 December 2026 apply to AI systems already in use?
Yes. From 10 December 2026, they apply to decisions made after commencement, even if the arrangement or the information came earlier (item 89).
Can a business use customer data to train an AI model?
Only with consent or another APP 6 basis. Where that cannot be clearly established, the OAIC’s Guidance on privacy and developing and training generative AI models (21 October 2024, updated 23 October 2024) says to seek consent, offer a meaningful and informed opt-out, or both.
Is publicly available personal information free to use for AI training?
Not automatically under the Privacy Act. The OAIC’s Guidance on privacy and developing and training generative AI models (21 October 2024, updated 23 October 2024) says publicly available or accessible data is not automatically lawful to use to train or fine-tune a model.
Does an AI project need a privacy impact assessment?
The OAIC’s Guidance on privacy and developing and training generative AI models (21 October 2024, updated 23 October 2024) recommends one, together with privacy by design.
What if an AI vendor stores data overseas?
If the vendor is outside Australia, a business bound by the APPs must, before disclosing personal information to it, take reasonable steps so that the vendor does not breach the APPs (APP 8.1). APP 1 is excluded from that duty. Where the APPs do not otherwise apply to the vendor, its act that would breach them is taken to be the business’s (section 16C).
Do small businesses have to comply?
Generally not, if annual turnover is $3,000,000 or less, unless an exception applies, such as disclosing personal information for a benefit (section 6D).
Do I need an AI privacy lawyer?
For an AI tool that handles customer information, or software that makes decisions about people, yes. An AI privacy lawyer can map each use against the APPs and draft the privacy policy wording required from 10 December 2026. A free 30-minute consultation will tell you where you stand.