AI Law in Australia: Legal Privilege, Copyright and Automated Decisions

This episode concludes the Legal Matters series on artificial intelligence and the law, covering three topics that affect every Australian who uses AI tools in daily life: whether typing a legal question into ChatGPT destroys your legal privilege, who actually owns the content AI generates under Australian copyright law, and what rights exist when an algorithm makes decisions about you. Caller James also raises a practical dispute about a door to door battery sale that illustrates just how much can go wrong without a written contract.

Andrew Bell from Bell & Senior Lawyers has spent the past several weeks exploring how Australian law is adapting, or in many cases struggling to adapt, to the rapid integration of AI into business, government, and everyday life. The central thread running through the series is that AI tools are not thinking machines. They are sophisticated statistical systems programmed to always produce an answer even when they have no reliable basis for one. That fundamental characteristic has serious and underappreciated legal consequences for ordinary Australians.

Key Topics Covered This Episode

• ChatGPT and legal privilege: The United States v Heppner ruling confirmed AI-generated documents carry no attorney-client privilege — Does Talking to ChatGPT Attract Legal Privilege in Australia?
• AI waiving existing privilege: Uploading a privileged document to a public AI tool may permanently destroy that protection — Can Uploading Documents to ChatGPT Destroy Legal Privilege?
• AI hallucination in Australian courts: Three lawyers referred to regulators after AI documents contained fabricated citations — AI Hallucination and Fake Case Citations
• AI and copyright: Entirely AI-generated content has no copyright protection under Australian law — Who Owns AI Generated Content in Australia?
• Automated decisions: New Privacy Act transparency obligations from 10 December 2026 — Your Rights When an Algorithm Makes a Decision About You
• James’s caller dispute: Verbal agreement, $500 deposit, supplier substituting a different product — Are Verbal Agreements Legally Binding in Australia? · Door to Door Sales and Your Cooling Off Rights

ChatGPT Cannot Be Your Lawyer: The Landmark Heppner Ruling

Legal professional privilege is one of the oldest and most fundamental protections in the common law.¹ When a client communicates with a qualified lawyer in confidence for the dominant purpose of obtaining legal advice, that communication is protected from compelled disclosure. The protection exists because the justice system depends entirely on clients being completely frank with their legal advisers.

The critical question in 2026 is whether the same protection applies when someone types a legal question into an AI chatbot rather than picking up the phone to a solicitor. A federal judge in the Southern District of New York answered that question definitively on 10 February 2026 in United States v Heppner.² The defendant, Bradley Heppner, faced fraud charges and had used Anthropic’s Claude AI tool, acting without any direction from his lawyers, to prepare documents outlining his defence strategy and his analysis of the government’s likely case. He later shared those documents with his defence counsel. Prosecutors sought to compel production of the 31 AI-generated documents. Judge Jed S Rakoff granted the application and refused any claim of privilege.

Judge Rakoff’s ruling rested on three findings. First, Claude is not a lawyer and no recognised privilege can exist without a trusting human relationship with a licensed professional who owes fiduciary duties and is subject to discipline. Second, the documents were not confidential because Heppner had communicated with a third-party platform and had consented to Anthropic’s privacy policy, which disclosed that user data could be shared with third parties including government regulatory authorities. Third, Heppner had not used Claude to obtain legal advice because he did so without any direction from counsel, and Claude itself explicitly disclaims any ability to provide legal advice and recommends consulting a qualified attorney.³

The Australian Position: Federal Court Warnings and Regulatory Referrals

While Heppner is an American decision, the Australian legal position is heading in exactly the same direction and in some respects has moved faster. The Federal Court of Australia issued Practice Note GPN-AI requiring all practitioners and parties to verify AI-generated citations independently, to disclose any AI use in document preparation, and to confirm that AI-generated propositions are supported by real authority.⁴ Paragraphs 4.13 and 4.14 of that Practice Note explicitly warn that inputting confidential or privileged information into a public AI tool may compromise confidentiality and waive privilege.

In December 2025, an Australian court referred three lawyers to their respective state regulators after AI-prepared court documents were found to contain inaccurate and erroneous citations. The court’s judgment flagged the risk that inputting draft documents into a public AI program could lead to privacy violations, breaches of obligations regarding subpoenaed materials, and waiver of legal professional privilege.⁵

Clayton Utz confirmed in April 2026 that the same principles apply under Australian law: legal professional privilege requires confidentiality as a key ingredient, conduct inconsistent with maintaining confidentiality may waive privilege objectively regardless of the user’s intent, and public AI tools threaten both the confidentiality and dominant-purpose requirements that Australian privilege law demands.⁶

The distinction that matters is this: when a qualified lawyer directs and supervises the use of AI as part of their advisory or litigation work, with proper confidentiality controls in place, Australian courts are more likely to regard the resulting communications as privileged. When an individual uses a public AI tool of their own accord without legal supervision, as Heppner did, no privilege arises and no existing privilege survives.

AI Hallucination: When the Machine Invents Cases That Do Not Exist

AI hallucination is the term used when a large language model generates entirely fabricated information with complete confidence. It has occurred with ABN lookups, business names, email addresses, and most dangerously in legal proceedings, court case citations that do not exist, citations attributed to the wrong courts, and legislative provisions that have been repealed or never existed.

In Australia, the December 2025 referral of three lawyers to regulators is a watershed moment.⁷ The Federal Court’s Practice Note GPN-AI makes clear that lawyers must verify the existence of every legal authority cited before filing any document, regardless of how that citation was produced.⁸ For self-represented litigants, the risk is equally serious: citing a non-existent case misleads the court, which may constitute contempt and can result in adverse cost orders or a strongly worded rebuke that damages the remainder of the case.

Andrew Bell noted on air that the same problem arises with legislation. AI tools regularly describe provisions from the wrong version of an Act, cite sections that have been amended, or apply American or United Kingdom law as though it were Australian law, often with no indication that the jurisdiction has changed. The output reads as authoritative while being substantively wrong in ways that are dangerous for anyone without the legal training to identify the error.

Who Owns AI Generated Content? Copyright and the Human Authorship Requirement

Copyright in Australia arises automatically the moment a human being creates an original work that reflects their independent intellectual effort.⁹ It requires no registration, no fee, and no formality. It protects literary works, artistic works, musical compositions, photographs, films, and other original material for 70 years after the calendar year in which the author dies.¹⁰ The owner of copyright has the exclusive right to reproduce, publish, and communicate the work to the public.

The difficulty with AI-generated content is that Australian copyright law has always required a human author.¹¹ Content generated entirely by an AI tool from a single prompt has no copyright protection because no human author has contributed the required independent intellectual effort. That means any competitor, any third party, or any member of the public who obtains access to that content can reproduce it, publish it on their own website, distribute it to their clients, or otherwise use it without infringing any copyright. The business that paid for the AI subscription and generated the content has no exclusive rights in it whatsoever.

Andrew Bell illustrated the human authorship principle on air with two analogies that reflect the correct legal logic. Courts in various jurisdictions have considered whether works produced by non-humans, including paintings attributed to elephants given a brush and photographs taken by animals operating cameras without human direction, can attract copyright. In each case the principle applied is the same: copyright vests in the human who makes the creative choices, not in the animal or the machine.¹² If a human presses the shutter, copyright belongs to that person. If the animal presses it unattended, no copyright exists in anyone. The same logic applies to AI: if a human author makes genuine creative choices and contributions beyond a single prompt, copyright may vest in them. If the AI generates the work autonomously, there is no copyright owner.

The practical implications for Gold Coast businesses are significant. Marketing copy, website content, product descriptions, blog articles, and social media posts produced entirely by a single AI prompt can be freely copied by any competitor. Businesses wishing to protect their content need either to produce it themselves with genuine human creative effort, or to ensure that a human author makes sufficient original contributions to the generation process to satisfy the independent intellectual effort requirement.

Automated Decisions: When an Algorithm Controls Your Life

The 2026 federal budget included a funding commitment to automate decisions made by Services Australia, the Department of Veterans Affairs, and other Commonwealth agencies using artificial intelligence, with the stated objective of reducing the human workforce required to process Centrelink claims, Medicare payments, NDIS eligibility determinations, and pension applications.

Andrew Bell described this development as deeply concerning for two reasons. First, large language models and AI decision systems do not produce deterministic results. The same input does not reliably produce the same output, meaning a system making eligibility decisions about pensioners or NDIS participants may produce inconsistent outcomes that have no rational basis and no human accountability. Second, the models underlying these systems have been trained predominantly on historical data drawn from English language sources. Historical data encodes historical biases. An automated decision system may be rejecting applications based on factors including surname, postcode, age, or other demographic characteristics that no human reviewer would consciously apply and that no member of the development team may have identified as a problem.

What the Law Currently Provides and What Is Changing

The most effective current avenue for challenging an automated government decision is to request the reasons for the decision in writing under the Administrative Decisions (Judicial Review) Act 1977 (Cth) or, in Queensland, the Judicial Review Act 1991 (Qld).¹³ If the automated system failed to take a relevant consideration into account, the decision may be susceptible to judicial review. As Andrew noted on air, this process is difficult, slow, and provides no guarantee of a different outcome.

A significant new protection is arriving. The Privacy and Other Legislation Amendment Act 2024 (Cth) introduces mandatory transparency obligations for any organisation, including government agencies, that uses personal information in automated or substantially automated decision making that could reasonably be expected to significantly affect a person’s rights or interests.¹⁴ These provisions come into force on 10 December 2026. From that date, organisations must update their privacy policies to disclose the kinds of personal information used, the types of decisions made by automated systems, the logic applied, and the most influential factors in those decisions. The Office of the Australian Information Commissioner has stated that “the algorithm decided” will not be an acceptable explanation.¹⁵

There remains no general right in Australia to require that an automated decision be reviewed by a human being. This right exists in several overseas jurisdictions and has been recommended by the Australian Human Rights Commission.¹⁶ It is expected to be legislated in Australia in coming years but has not yet been enacted.

Caller: James and the Door to Door Battery Dispute

James called in with a dispute that began when salespeople arrived uninvited at his door selling batteries. He agreed verbally to purchase a 10 watt battery for $500, paid a deposit, and signed nothing. The supplier subsequently communicated only by email, which went to James’s junk folder. The supplier is now attempting to substitute a 9.3 watt battery instead of the agreed 10 watt product and is demanding full payment in advance before commencing installation.

Andrew Bell’s advice addressed two distinct legal issues.

The verbal contract. The absence of a signed document does not mean there is no contract. James made a verbal agreement with sufficiently certain terms, being a specific product at a specific price, and paid a deposit. That combination is almost certainly sufficient to constitute a binding contract.¹⁷ The supplier is now proposing to deliver something different from what was agreed, which is a breach of contract. James’s options are to accept the substituted product by negotiating a variation, or to treat the supplier’s inability to deliver the agreed product as a frustration of the contract and demand a full refund of the deposit.

The demand for full upfront payment. While requiring upfront payment is not inherently a sign of fraud, it is a legitimate concern in the context of a door to door sale where the supplier had already been difficult to contact. Andrew’s practical recommendation was direct: request the deposit back and engage a different provider in what is a competitive market.

Key Takeaways

1. AI conversations are not protected by legal professional privilege in Australia. Judge Rakoff in United States v Heppner confirmed that AI tools are not lawyers, communications with them are not confidential, and no privilege arises. Australian courts and the Federal Court’s Practice Note GPN-AI are aligned with this position.

2. Uploading privileged documents to a public AI tool may permanently waive privilege. Waiver of legal professional privilege in Australia is assessed objectively under Mann v Carnell (1999) 201 CLR 1. Conduct inconsistent with maintaining confidentiality destroys the protection regardless of what the user intended.

3. AI-generated case citations must be verified on AustLII before any use. Three Australian lawyers faced regulators in late 2025 for submitting AI-prepared documents with fabricated citations. Every citation must be independently confirmed before filing or relying on it.

4. AI-generated content has no copyright protection in Australia. Copyright requires a human author who contributes independent intellectual effort under IceTV. Content produced entirely by a single AI prompt can be freely copied by any competitor or third party.

5. New automated decision transparency rights commence 10 December 2026. The Privacy and Other Legislation Amendment Act 2024 (Cth) will require organisations to disclose the logic, data, and key factors behind automated decisions. “The algorithm decided” will not be acceptable.

6. A verbal agreement plus a deposit is a binding contract in Queensland. If a supplier cannot deliver what was agreed, they have frustrated the contract and you are entitled to a refund of the deposit. The absence of a signed document does not eliminate the agreement.

7. Door to door sales attract a 10 business day cooling off period under Australian Consumer Law. This right cannot be contracted out of and applies to any unsolicited agreement entered into at a consumer’s home.

8. AI tools carry no warranty for accuracy and there is currently no legal recourse in Australia. Every major AI provider expressly disclaims liability for the accuracy of outputs. There is no successful Australian precedent awarding damages against an AI company for inaccurate information.

9. Algorithmic bias is a real and unresolved risk in Australian government decisions. AI systems trained on historical data may discriminate by surname, postcode, age, or other factors in ways that are very difficult to detect or challenge under current Australian law.

10. Enterprise AI used under legal supervision is safer for privilege, but not risk-free. If a qualified lawyer directs and supervises AI use with proper confidentiality controls, Australian courts are more likely to treat resulting communications as privileged. Public AI tools used without legal supervision provide no such protection.

Listener FAQ: Deep Dive Articles

The following resources expand on the legal topics discussed in this episode. Each article provides a detailed explanation with relevant Australian case law and legislation.

Related Episodes

This Technology Law Series

• Episode 16: Technology Law Part 1 — Business, Cyber and the Digital Economy — internet law, safe harbour provisions, DMCA, cyber security obligations for Queensland businesses
• Episode 17: Technology Law Part 2 — Consumer AI, Safe Harbour and Internet Law — what AI actually is, how LLMs work, consumer rights in the digital economy

Related Blog Articles

• Legal Risks of AI for Business – Australia, US and UK — the definitive guide comparing Australian, US and UK AI regulatory frameworks, with a practical checklist and December 2026 compliance deadlines

Also Relevant

• Episodes 13 and 14: Administrative Law — the administrative law framework for challenging government decisions, directly applicable to automated Centrelink and NDIS decisions
• Episode 11: Debt Collection and EPOA — recovering deposits paid to non-performing contractors
• Episode 12: Loan Agreements and Debt Collection — contractual obligations and consequences of failure to perform
• Episode 7: QCAT and Disputes — practical process for resolving consumer disputes including deposit recovery

Next Week

Andrew flagged that the AI law series may be set aside for a different topic next week, depending on listener demand. Colin and Andrew have further AI material prepared and may return to it if listeners call in with questions. Tune in to 4CRB every Tuesday at 9am or visit bellsenior.com.au for resources and show notes published during the week.

Annotated Transcript

Segment 1 — Introduction & AI Law Recap 📎 See also: AI Legal Privilege Australia

Announcer: 4CRB now presents Legal Matters, proudly brought to you by Bell and Senior Lawyers. Call them today for all your legal needs on 07 5532 8777. This program provides general legal information only. It’s not personal legal advice. Everyone’s situation is different, so please seek independent advice for your own circumstances.

Colin Balewski: Well, it’s a very good morning to Andrew Bell from Bell and Senior Lawyers. Good morning Andrew.

Andrew Bell: Good morning Colin. Thanks for having me back yet again talking more about technology law.

Colin Balewski: We are indeed. Well this is your show, it is Legal Matters. This is your opportunity this morning to get some free legal advice live on air. We’d love to hear from you this morning. Do not leave it to the end of the program, 07 5520 8888 is the number. 5520 8888. And we’d love to hear from you this morning. Last couple of weeks we’ve been talking AI law and we’ve had some great feedback from people on the topic. We’ve enjoyed, I’ve enjoyed the conversation because it’s what I studied way back when and uh this is a very interesting topic as to where the law sits in all of this or where it’s potentially going to sit if it hasn’t made up its mind yet. Maybe we’ll start with that recap Andrew as to what we discussed in the last couple of weeks.

Andrew Bell: Sounds good. Yeah, I agree. AI is now being woven into every part of everything that we do. So while it wasn’t a big consideration a few years ago, it really is becoming important that we understand what we can and can’t do. So two weeks ago we talked about internet law and why Google and Facebook etc can host what they can host and seemingly get away with it. The… these were the safe harbor provisions that came from America through the Digital Millennium Copyright Act or DMCA, and then got incorporated into Australian law via the US free trade agreement. And then last week we started talking about what AI actually is. And my central point was that large language models aren’t thinking machines, they’re really sophisticated auto completes. So they look at the statistically most likely word that comes next and they’re programmed to always produce an answer even when they genuinely don’t have one behind the scenes, which can create all sorts of chaos.

Segment 2 — Legal Privilege & ChatGPT (United States v Heppner) 📎 See also: AI Waiving Legal Privilege

Colin Balewski: Absolutely. Well this is your program. 07 5520 8888 is the number. We are talking AI law this morning, but that doesn’t preclude any legal question you may like to pose this morning. 07 5520 8888. Andrew, prior to us going on air, we talked about a case decided in New York in February this year that you said every single person who has ever typed a legal question into ChatGPT needs to hear.

Andrew Bell: Exactly. So Colin, in Australia and in America, if you go and see a lawyer and tell them that you might have a problem, that conversation is protected. We call it legal professional privilege. So your lawyer can’t be forced to reveal what you’ve told them, and it’s one of the oldest and most fundamental protections in our legal system. And it exists because people need to be completely open and honest with their lawyers. Um, however, what happens if you’re treating something like your lawyer, but they’re actually an AI robot? Okay, this is an interesting one. So in February this year a federal judge in New York City handed down a decision in a case called the United States against Heptner. And in this case, the defendant had used a similar program to ChatGPT called Claude to research and work through his own legal position. He later picked up those documents and shared them with his defense lawyers. Prosecutors then tried to get a subpoena, which means that Claude was forced to hand over the information that they had about him. And that was granted by the judge. The defense argued that they were protected documents because they were relating to his legal defense and that he had shared them with counsel, and the judge rejected it. He said the documents went straight to, there is no basis for any claim of attorney-client privilege. And the documents went straight to the prosecutors. So the defendant’s own AI research was used against him in court.

Colin Balewski: Okay, so let’s be really clear about this. The moment you type a legal problem or a legal question into ChatGPT instead of ringing a solicitor or a lawyer, your protection is gone.

Andrew Bell: Well that’s definitely the case in America. So yeah, if this now this does exist with other online platforms already. If you Google things, I believe the Baden-Clay example and there’s been a number of other high-profile murders and things where people have Googled things and it’s been used against them. However people treat their AI like it’s a person and they talk to them and trust them, and they tell them things that they probably wouldn’t tell anyone else. And now it turns out they definitely shouldn’t be telling anyone else.

Colin Balewski: Yeah, okay. That’s something that we hadn’t considered because I notice every day we ask ChatGPT or any of the AI bots out there more and more extensive questions surrounding business, surrounding personal efforts. It does a lot for you, but you’ve got to be careful about what it is that you’re discussing, even if it is a legal problem because it may end up being something that’s held against you.

Andrew Bell: It may. I’d treat it like a postcard. Don’t put anything into it that you don’t want anyone else to be able to read on the other side.

Segment 3 — AI Hallucination & Fake Cases 📎 See also: AI Hallucination Courts Australia

Colin Balewski: All right, this is your program. 07 5520 8888. If you’d like to pose a question to Andrew Bell from Bell and Senior Lawyers this morning, and do not leave it to the end of the program, we’d love to hear from you. All right, moving right along, we’re talking AI law and oh dear, oh dear Andrew, when I start to think about what you put forward there, um there’s lots of things that you could be asking. You could be asking about a contract. I’m thinking of selling something, I’m thinking of willing something. All of these protections, or all of these… and again we also have that issue that we talked about last week, is it giving you the correct advice?

Andrew Bell: Well it’s it’s very difficult to tell if it’s giving you the correct advice Colin, because it’s programmed to give you what… because it doesn’t know anything and it doesn’t independently check things… although some services are starting to do that, it’s just giving you what it thinks the most likely answer is going to be. Um it comes up all the time, particularly in more narrow fields. It may give you an answer for Australian law, for example, and says it’s Australian law, but it’s actually American law. And our laws differ substantially in some areas and they’re very similar in some areas. Um yeah, what you put into it and what you get out the other end, you need to have the skills to be able to validate that it’s correct and look at it with a very keen eye. And even though you normally trust it and it might be 95% right, that 5% could be really detrimental to you.

Colin Balewski: So you need to check it for reasonableness is probably what you’re saying. Maybe the practical message in all of this is use AI, but make sure that you’ve then gone and done your homework afterwards, maybe it’s a good starting point. But again, don’t share with it things that you wouldn’t be happy if someone else was to put it up um and display it in front of your boss, your wife, your colleagues, etc. All right, for the last couple of weeks we have been discussing this topic, but we would love to hear from you this morning. 07 5520 8888 is the number. Pose any legal question you like and have it answered live on air. So for the last couple of weeks we have been talking AI and it invents legal cases that do not exist. Has this actually happened?

Andrew Bell: So yes, it’s happened overseas and it’s happened in Australia. So what we were saying last week Colin, you used the example of searching for an ABN on ChatGPT and it told you very confidently the business’s ABN, but when you went to look for it, the ABN didn’t exist. And the same thing happens with Australian cases. So there’s been a number of times where lawyers have got themselves in serious trouble, and my understanding is that self-represented people are facing the same problem, they’re turning up to court quoting a case that doesn’t exist. Now that is worse than losing a case, because you are misleading the court which is contempt. You have an obligation to be completely frank with the court and you’ve obviously put no effort into ascertaining whether that actually existed. Now a lawyer can be struck off and lose their practicing certificate for doing that, and if you’re self-represented you may receive the wrath of the judge, adverse cost orders, or even further sanction. So, this is called hallucination where the AI completely makes things up, and if it makes up a quote or an ABN or an email, that’s one thing. If it makes up a completely fake High Court case and relies on it, you’re in a whole lot of trouble.

Colin Balewski: Okay, that is uh a trap for young players. We have heard a lot of these stories, um and some of them are quite strange indeed. But I think it comes down to the bottom line Andrew, I one of the early ones as uh AI was coming out, someone said, write me a a contract uh to sell a product from one to the other. Um and protect my business and so forth. And then after the AI had generated it, he took it to a solicitor and said check this for me and tell me what you think. The solicitor came back and said, yeah, great, probably couldn’t have written it any better than what this one has, there’s probably a couple of clauses I could add. But when it goes wrong, who are you going to turn to? You can’t ring AI or talk to AI and have him solve that problem. And then where do you stand when you turn up at a solicitor’s door and say, hey I wrote my own contract, and now I’m in trouble?

Andrew Bell: Well unfortunately, you have nothing to fall back on. They all say, there’s if you read the terms and conditions that most people don’t, they all turn around and say that they have no warranty for any of this stuff that they produce. Um, it is being tested at the moment because people are suing AI companies um particularly in the states related to mass shootings or other things that it’s encouraged them to do, but there is no comeback. You use AI and you get what you get, it’s similar that if you go to some random website and copy information off it and then rely upon it. If you’ve got no contractual relationship with those people, you probably have very few grounds to do anything about it.

Colin Balewski: And that’s uh that’s probably coming back to one of those golden rules, it’s all good until it’s not. And you need to have backup when those things happen. We might go to a word from our sponsor, but before we do, uh we’d love to hear from you. 15 minutes left in the program. This is your opportunity to get some free legal advice live on air. Doesn’t have to be to do with what we’re talking about today, it can be any topic you like. 07 5520 8888.

Announcer: Wills, probate, powers of attorney, or property matters are legal issues that can feel overwhelming. Don’t put off dealing with them. Call our trusted sponsors, Bell and Senior Lawyers. They explain everything in plain English and guide you through every step of the way. For caring, experienced advice, phone 07 5532 8777 or head to bellsenior.com.au today. Bell and Senior Lawyers, making legal matters simple.

Segment 4 — Copyright and AI Generated Images/Content 📎 See also: AI Copyright Australia

Colin Balewski: All right, we might move on to AI and copyright now, uh, and this is a a very broad discussion. I know that it’s been followed uh very closely by not only a number of uh my friends, but a lot of the volunteers have been talking about it around the uh the table. Uh all right. A number of Gold Coast business owners will be interested in this one. Who owns the generated copyright and images? Before we get into AI specifically, can we explain what copyright actually is for listeners uh who are not familiar with it and then we’ll dive into the uh the minutiae, let’s call it.

Andrew Bell: Sounds good. Well yeah, copyright um is very important and AI is a great tool to generate ideas, and I’m not saying don’t use AI, I’m just saying we need to be careful what we do use it for. But copyright is a form of intellectual property and it gets generated automatically the moment you create something original. In Australia we don’t need to register copyrights, so you might be familiar with the ‘C in a circle’ you’ll see on stuff, in other countries you do need to register it. You don’t need to pay for it, and it just exists for about 70 years. Literary works, artistic works, music, pictures… so all the sorts of things you might want to put on a website, Colin.

Colin Balewski: Yes.

Andrew Bell: And so that’s original information and that needs to be legally protected. So the creator of the work owns the copyright, that means you have the exclusive right to reproduce it, publish it, or communicate it to the public. Uh if a business puts an original photo on their website, they own the copyright because they created it, and you can assign those rights, and you can also protect other people from using them. So copyright is a very valuable tool for businesses to make sure that what they put out there into the world, that they’ve got some control over. Now, where AI comes into the picture, um might shock people. But a generated a document that’s generated entirely by Claude or ChatGPT from a single prompt, under Australian law is not protected by copyright. So anyone can copy it freely if they can get access to it. And this is completely different to things that you’ve written yourself. You own it because you’re the human author. The copyright law requires a human author in Australia, not a machine. If AI writes an entire report for your business, your competitor can lift it, publish it on their site, or you know, print it out and give it to their clients, and under copyright law there isn’t a lot you can do about it.

Colin Balewski: So that’s the minutiae. Does that include AI generated photos and images?

Andrew Bell: Yes, there’s a there’s a number of cases that that have gone through quite some substantial courts that that relate to copyright pre-AI where it was determined that machines and animals can’t own copyright. So there is a a famous case relating to elephants that were creating art, given a paintbrush, let them choose the colors, people pay for that art. And then other people were reproducing it and they got sued saying we own the copyright to this. Well, it was determined that it wasn’t an original piece of art because it wasn’t created by a human and therefore there was no copyright. There’s been similar cases about monkeys taking photos with with a camera. And again, the monkey can’t have the copyright, but if I press the button the copyright would vest in me. So yeah, you’ve got to be really careful depending on on what you’re doing, if you’re going to rely on that for anything major.

Segment 5 — Automated Decisions by Algorithms 📎 See also: Automated Decisions Australia

Colin Balewski: All right, we might move on to another component of AI, but first again, 8 minutes left in the program. 07 5520 8888, do not leave it any longer. Ring now to ask your question or legal question of Andrew this morning. Uh, when AI makes decisions about you… Now I know we had a caller a couple of weeks ago that posed something similar to this, their bank used an algorithm to refuse their home loan with no explanation. What can they do and is the law changing in this area?

Andrew Bell: So the law in this area is going to change fast, but we don’t exactly know how it’s going to change. But um, AI is able to make decisions and is being used to make decisions already. The most recent budget handed down in May has taken that very seriously. The government committed millions of dollars to grants to um make sure that Services Australia, who do Centrelink and Medicare, Department of Veterans Affairs, and a number of other departments, to start automating decisions made around people using AI so they can reduce their human workforce. Now I find this very concerning. Um, last week we talked a little bit about how these large language models and AI don’t create what’s called deterministic results. You can put the same info in and get a different output out the other side. So the problem is, if they’re making a decision whether you’re eligible for NDIS, whether you’re eligible for a pension, whether they want to cancel or give you something that you’ve applied for, and then they reject it, on what grounds are they doing so? Now AI models take in lots of large amounts of information and produce an answer out the other side, but how do we actually know why they made that decision? Um, at the moment, all you can really do, if we go back to the weeks we discussed about administrative law, is ask for the reasons for the decision. And if they didn’t take everything into account that they should have taken into account, you might be able to get it reversed. But that’s going to be quite a difficult long process, and these decisions are going to be automated all the time.

Colin Balewski: One of my favorite things to do with these types of situations, whether it’s call centres overseas or whether it is automating processes or uh utilising software to make deterministic solutions is that while it may be faster and it may be uh easier for the company that is selling you the product or service, it really isn’t great for us. We end up spending a lot more time in involved in either getting that product or service or being completely frustrated by it, and hence some of the uh the Centrelink call uh numbers where you’re forever being told your call is important to us, please wait, but meant to be more uh more streamlined and efficient and it not and it isn’t.

Andrew Bell: Absolutely. And overseas some people at least have the right for a human to review an automated decision, we don’t have that here. It’s something that’s been recommended, and I think we’re going to find some things go horribly wrong before that then comes in here as well. Um, but as someone who’s built systems to make large amounts of decisions around approving credit, turning off people’s mobile phones and all sorts of other bits and pieces, there’s only so much information that you can take on board. And um how these decisions work is very difficult to actually describe to someone. But if you see the person next to you get approved and you get knocked back, why is it? Now I also think that discrimination and other things are going to come into it as well. A lot of these large language models have been trained on English language systems, trained on historical data. So it may have rejected you because of your surname, or because of your age, or other perceived biases that are built into the model that no one even is aware that they exist. So this is a rapidly developing area of law and not one where there’s a lot of actual guidance that the law gives us.

### Caller 1 — James | Door to Door Battery Sales 📎 See also: Verbal Contracts Australia & Door to Door Sales

Colin Balewski: We’ve got our first caller, it is a very good morning to James. Ask your question of Andrew, James.

James (Caller): Good morning Andrew, how are you?

Andrew Bell: I’m good thank you James, how are you?

James (Caller): Yeah, great. Now um, I just want to know, I um had some people come in off the street uh selling batteries. Mm-hmm. Now now um, we spoke to them and we probably didn’t do our homework properly, but anyway um, we organized to buy a battery from them. Now we never signed nothing, um we tried to get in touch with them uh since. Mm-hmm. Now what’s happened is uh they don’t answer their phone, um anyway we managed to get in touch with them just recently and they said that um they sent us an email. Now we didn’t receive the email, although we thought we didn’t, but then it was in the junk mail, so we did receive the email. But we paid them $500 um for the battery, for a 10-watt battery. Now they want to give us a a 9.3 battery instead of a 10-watt. Mm-hmm. Um and we’re we’re really not happy with that. Um we ordered, now we never signed nothing to these blokes, so we ordered a 10 uh watt battery and that’s what we want. So how do we stand legally?

Andrew Bell: Well you keep saying you didn’t sign anything, but you did agree verbally and you have paid them some some money, so normally it would be considered that you have actually entered into a contract. You’ve gone beyond the negotiation stage. If it was sufficiently certain around what you’re ordering, and they can’t provide that, then you’d typically say that they’ve frustrated the contract and you should be able to get any deposit you paid back. You either get what you agree that they’re going to give you, or you negotiate a change to the contract. Um, but if you don’t want to negotiate the change to the contract, you can give them an the ultimatum. You either need to do what you said you’re going to do or you need to return my deposit.

James (Caller): Yeah, well these people also want the money up front before they do the job. Mm-hmm. And I’m I’ve been caught before by people where we’ve paid the money and we’ve we’ve just never got the job done.

Andrew Bell: Well I mean, anything can be negotiated in a contract. You can do it either way around, it doesn’t necessarily mean they’re a scammer. But I did also recently put in batteries and only paid after it was complete to make sure that it was um done to a high enough standard and it was signed off um, and the solar rebates came through. But again, you that this is to do with what you’ve agreed. So if you want to give them if you’ve already paid a deposit, I’d be questioning why you need to pay the rest of it up front. James, did you get any paperwork at all um describing uh what they were promising?

James (Caller): Yes, we did yes, we did get some uh paperwork what they were promising and they keep they just keep sending us an email now um, we need this money three or four days before we do the job. And I sort of think, well you know like, why should I have to pay up front to do a job um when the job mightn’t even be done or mightn’t be done properly?

Andrew Bell: I I think you’re right, but it does depend if there was any written contract or terms and conditions that you agreed to that said it would be paid up front. I can see why they may be asking for that, but it’s a very competitive market, I’d be recommending that you just ask for your money back and go and and go and speak to another provider.

James (Caller): Thank you very much Andrew. I certainly appreciate your time and what a great program you’ve got there.

Colin Balewski: Thank you James.

James (Caller): All the best. Thank you, bye.

Segment 6 — Closing

Colin Balewski: All right, well that went very very quickly and I noticed we’ve got a text message there as well which I can’t read across the room so we’ll do that off air. We didn’t get to the end of our program again. But that’s okay, we do enjoy your phone calls. Thank you to James who called in today. We would like to hear from more of you next week. And uh Andrew, thank you very much again for your time this morning.

Andrew Bell: Thank you Colin. I think we might might shelve we’ve got more AI stuff, if people pop bring it back by popular demand we might, but we might move on to a different topic next week.

Colin Balewski: I’m excited by uh the questions we didn’t get to, but we’ll negotiate that out Andrew. All right.

Andrew Bell: Thank you.

Colin Balewski: All right, thank you very much for your time this morning. This was Legal Matters and it is heard every Tuesday morning at 9am. And uh you can call in at that time, same time next week.

Announcer: You’ve been listening to Legal Matters, proudly brought to you by our trusted sponsor Bell and Senior Lawyers. Call them today for all your legal needs on 07 5532 8777. A copy of this and past programs and all relevant resources will be available for you to replay or download on bellsenior.com.au and 4crb.com.

Footnotes