Summary
- Businesses using AI must comply with the Australian Privacy Principles and the Privacy Act when processing personal or sensitive data, and should de-identify information, avoid inputting confidential data, and ensure transparency with customers around data handling and consent.
- AI-generated outputs are unlikely to attract copyright protection in Australia, as AI lacks legal personhood and the outputs may not satisfy the requirement for independent intellectual effort, meaning businesses should review the IP terms of any AI service provider before use.
- AI outputs can contain errors, hallucinations, and biased content reflecting prejudices in training data, meaning businesses must fact-check outputs and avoid relying on AI-generated content without first verifying its accuracy, fairness, and reliability.
- This article is a guide to the legal risks of using AI for business owners operating in Australia, produced by LegalVision, a commercial law firm.
- LegalVision specialises in advising clients on technology law and AI-related legal compliance.
Tips for Businesses
Review the IP and data terms of any AI tool before use. De-identify personal information before inputting it and disable settings that allow data to train AI models. Always fact-check AI outputs, particularly where legislation or legal analysis is involved, as AI models can reference non-existent laws.
Artificial intelligence (AI) offers businesses real opportunities to automate tasks, cut costs, and improve efficiency, but it also introduces significant legal risks that you cannot afford to ignore. Understanding where those risks lie is the first step to using AI responsibly and lawfully. This article explores the main areas of legal concern for a business when using AI, including areas such as data privacy and protection, intellectual property, discrimination issues, and liability.
Data Privacy and Protection
The use of AI for businesses usually involves processing large amounts of personal or sensitive data. Consequently, this can raise privacy concerns around the collection, usage, storage and disclosure of this information. Depending on the scope and nature of your business, you may be required under the Australian Privacy Principles (APPs) and Privacy Act to take measures to safeguard data. Even in the absence of statutory obligations, you should apply best practice principles. Additionally, you should have mechanisms in place to ensure transparency with customers around data handling and consent.
If your business uses generative AI within its operations, particularly when dealing with confidential, personal, or sensitive information, you should:
- avoid inputting confidential information;
- ensure the information provided is thoroughly de-identified to mask personal information (de-identification is important in lowering risks associated with data breaches and unauthorised disclosures and involves removing or altering details that may reveal the identity of individuals); and
- select the option where any data inputted is not to be used to train the AI model.
Some industries require a customer’s informed consent before they input their data into an AI model. For example, in the health industry, doctors must obtain informed consent from their patients before inputting patient notes and data into a transcription software service. They must also de-identify the customer’s information.
Intellectual Property Rights
When using AI for your business, you must consider who owns the intellectual property (IP) created by the AI or inputted into the AI model. You may be familiar with the problems faced by several AI companies who have trained their AI models using copyrighted materials without authorisation from the authors or owners. This means that you should be careful in what you use as output from an AI model. For instance, you could be infringing someone else’s copyright if you rehash the AI-generated work product instead of using the output for inspiration or guidance.
Ownership of Inputs
Every AI service provider has its own policies or terms on IP. Acquaint yourself with these terms before using their software. Typically, human-generated inputs are protected by copyright law.
Ownership of Outputs
AI lacks legal personhood and is incapable of owning any outputs that it generates. Under the Copyright Act 1968 (Cth), copyright remains in an original work where the author is a ‘qualified person’; that is, an Australian citizen or resident. The scope of copyright protection can be limited if a human author has not infused the work with a distinct ‘creative spark’ or exerted ‘independent intellectual effort’ or ‘skill and judgment’. While the position in Australia is not settled, it is unlikely that copyright protections will apply to AI-generated outputs.
Continue reading this article below the formCall 1300 544 755 for urgent assistance.
Otherwise, complete this form, and we will contact you within one business day.
Accuracy and Reliability of AI Outputs
You must exercise caution when relying on AI-generated outputs as they may contain errors and inaccuracies. Always exercise due diligence and fact-check the information it produces. This way, you can be certain you are not relying on false and unreliable outputs. AI models can also have shortcomings in their ability to analyse complex information such as legislation. Moreover, it might even reference non-existent laws and regulations (this is known as ‘hallucination’).
Discrimination and Bias
AI-generated content has the potential to reflect biases inherent in the training data it relies upon, leading to biased output. Suppose the AI model is trained on data that is characterised by prejudices, stereotypes, small sample sizes and limited diversity. Accordingly, the AI model is highly likely to mirror those prejudices and stereotypes in its work product. To minimise generating results that discriminate against certain groups, ensure you train your AI models on wide-ranging and unbiased data. You should avoid blindly relying upon AI output without first ensuring its quality, fairness and accuracy.
This fact sheet outlines your rights and obligations as an AI artist regarding intellectual property and copyright.
Key Takeaways
AI has great potential to transform your business, but with its use, several legal concerns arise. As a business owner, you must be well-informed and diligent when navigating the intricate legal landscape surrounding AI. This will allow you to mitigate risks effectively.
If you need guidance on understanding and fulfilling your legal obligations when integrating AI into your operations, LegalVision provides ongoing legal support for businesses through our fixed-fee legal membership. Our experienced AI lawyers help businesses manage contracts, employment law, disputes, intellectual property, and more, with unlimited access to specialist lawyers for a fixed monthly fee. To learn more about LegalVision’s legal membership, call 1300 544 755 or visit our membership page.
Frequently Asked Questions
Businesses should comply with the Australian Privacy Principles and the Privacy Act by safeguarding data. It is important to avoid inputting confidential information into AI systems, thoroughly de-identify personal information, and choose settings that prevent the use of input data for training AI models. Transparency with customers and obtaining informed consent, especially in sensitive industries like healthcare, are also crucial steps.
Businesses must consider who owns the intellectual property related to AI, both in terms of inputs and outputs. Human-generated inputs generally retain copyright protection, while AI-generated outputs may not, as AI lacks legal personhood. Businesses should familiarise themselves with the intellectual property terms of AI service providers in order to avoid infringing on copyrights when using AI output.
Yes. AI models trained on biased or unrepresentative data can produce discriminatory outputs that reflect prejudices and stereotypes. To minimise this risk, train AI models on wide-ranging, diverse, and unbiased data, and always review outputs for fairness and accuracy before relying on them.
AI models can produce inaccurate outputs, including references to non-existent laws and regulations, known as ‘hallucination’. Always fact-check AI-generated legal information and exercise due diligence, as AI has limitations in analysing complex materials such as legislation.
We appreciate your feedback – your submission has been successfully received.
