If you use a generative AI tool in your everyday work activities, you’re unlikely to be thinking– does this breach copyright?
It’s easy to presume that AI content is inherently “safe” to use. However, in practice, that is not necessarily the case.
Copyright risk does not disappear with AI
New Zealand’s Copyright Act 1994 recognises copyright in computer‑generated works. This means that, in many cases, copyright can continue to apply in material produced using AI.
However, the key issue is not simply who owns the output. It is whether that output incorporates, reproduces, or is derived from someone else’s protected work.
At both the AI training or input stage and the AI output stage, there is a risk of copyrighted materials being substantially reproduced without authorisation from its creator or author. There is currently a large gap with how businesses identify and handle these risks in their use of AI tools, mostly due to the speed at which society has adopted such technologies.
AI training and inputs
Generative AI tools do not ‘create’ in isolation. They are trained on large datasets containing text, images, code, and other material, often sourced from across the internet. Where that material includes copyrighted works, and no statutory exception applies, there is ongoing uncertainty as to whether infringement may occur at the training stage.
On the face of it, uploading or otherwise feeding a copyrighted image into an AI tool generally creates at least one digital copy of that image. “Copying” in New Zealand includes reproducing or storing a work “in any material form (including any digital format), in any medium and by any means.” Making such a copy, without the copyright owner’s permission and outside any exception, infringes the owner’s exclusive right to copy.
While some exceptions exist for incidental inclusion or fair dealing for research/private study, unlike jurisdictions such as the United States, New Zealand does not have a broad “fair use” doctrine. Our “fair dealing” exceptions are more limited in scope and are unlikely to extend to commercial AI training activities.
So, while manipulating photos on Chat GPT may be an enjoyable pastime, a workplace’s internal policies regarding inputting images into generative AI should look to ensure employees are aware of some key considerations. Employees should only use images that:
- the business owns or is licensed to use.
- the business has express permission to use for the specific purpose.
- are in the public domain.
Training and awareness programmes with employees, could start with making sure employees use appropriate prompts to request original concepts rather than reproductions of existing works. Avoiding logos or well-known characters is also a good idea.
The same applies for text or works not owned by the business, but that employees are using to feed information into the AI tool for context or background. This act would still amount to copying at the time the information was inputted into the AI tool.
The real exposure: AI outputs
For most organisations, the more immediate risk arises at the point of use.
AI generated outputs can, in some circumstances, be substantially similar to existing copyrighted works. This is particularly relevant for creative outputs such as written material, visual content, and code.
For images, if the output reproduces distinctive, protectable elements of the source image (composition, characters, distinctive expression) to a substantial extent, using or sharing that output can infringe the copyright owner’s reproduction right.
If that output is then published, shared, or used commercially, there is a risk of copyright infringement.
Importantly, the fact that the material was generated by AI and that the user may have been unaware of the similarity, will not necessarily provide a defence.
Three ways to reduce risk
There are some practical ways workplaces can look to reduce this risk:
- Review platform terms carefully and understand your provider’s approach
AI providers commonly assign ownership of outputs to users, while disclaiming liability for infringement. Responsibility for how outputs are used is often shifted to the user too.
In practice, the risk allocation depends heavily on the specific agreement’s terms. A few AI providers offer very narrow indemnities for generative AI outputs on the basis that specific guardrails are adhered to. There may also be other conditions such as liability caps and how the agreement defines “authorised use.” Many agreements also place obligations on the subscriber to implement acceptable use policies internally, effectively pushing compliance responsibility to the organisational level rather than the individual user.
In most cases, the organisation accepts the terms of use and is responsible for ensuring employees comply with them. Knowing where the risk sits contractually is important for businesses to be aware of.
- Implement a reasonableness check to generative AI outputs
For AI generated content being used externally or commercially, businesses should apply more scrutiny to the output and have mechanisms for checking the work isn’t a derivative of existing copyrighted works.
There are a variety of technological tools designed to compare outputs against existing content such as plagiarism detection tools, search-engine checks and image similarity detection tools which could be appropriate for high-risk publications. Otherwise, a human review and approval process may be sufficient.
- Document human input and decision‑making
Maintaining records of prompts, edits, and creative direction can help demonstrate independent creation and strengthen ownership positions. This could form part of the sign-off process when reviewing the outputs.
What happens if it is a breach?
The usual primary and secondary infringement provisions apply (e.g., copying, issuing copies to the public, communication to the public, and providing means for making infringing copies). Any AI related claim would be evaluated through these existing sections of the Copyright Act 1994, and a breach may give rise to civil penalties, including damages, account of profits, and injunctive relief.
Key takeaway
The legal framework in New Zealand is still evolving in response to generative AI. The Government has indicated that further work is underway to consider how copyright law should apply in this context.
In the meantime, existing copyright principles continue to apply and AI tools should be treated like any other third‑party input application. Where the origins of content are uncertain, and where outputs may be derived from existing works, a degree of caution is warranted, particularly when used commercially.
