Comment: For New Zealand businesses and content creators, there is a growing need to understand how copyright law applies to works created using generative artificial intelligence.<\/p>\n
Under New Zealand\u2019s Copyright Act 1994, copyright subsists in \u201coriginal works\u201d of various types, including literary, dramatic, musical and artistic works. The requirement for\u00a0originality\u00a0is central. The works must originate from the author and not be a copy of, or otherwise infringe, another person\u2019s work.<\/p>\n
The Act was drafted more than 30 years ago and does not expressly address works created by GenAI. However, the categories of works covered by the Act are broad and, in theory, can capture many typical outputs created with the assistance of GenAI (for example text, images, tables, or compilations). Whether those outputs are \u201coriginal\u201d for the purposes of the Act is less clear.\u00a0<\/p>\n
While the issue remains untested, it is likely this will depend on the degree of human skill, labour, and judgement applied to create the work \u2013 for example, the effort applied in developing prompts and refining the GenAI output. In simple terms, the higher the degree of human involvement in the creative process, the more likely the resulting work will be considered original and protected by copyright.<\/p>\n
The originality threshold will depend heavily on the facts of each case, but might apply where, for example, a marketing creative designs a campaign and develops their ideas, sketches and draft text using GenAI to improve the campaign iteratively. The threshold is less likely to be met where someone relies on an unedited output produced by a single basic prompt. In cases of doubt, to minimise the prospect of disputes over whether copyright subsists in works created with the assistance of GenAI, businesses should carefully record the creative process and human input involved to help support claims of originality.<\/p>\n
\u2018Computer-generated works\u2019\u00a0<\/p>\n
The Act also recognises copyright in \u201ccomputer-generated\u201d works, defined as works generated \u201cin circumstances such that there is no human author of the work\u201d. For such works, the \u201cauthor\u201d is deemed to be the person \u201cby whom the arrangements necessary for the creation of the work are undertaken\u201d.<\/p>\n
The scope of this provision is uncertain. It applies only to works that are \u201coriginal\u201d, but originality generally requires the expression of an author\u2019s creative choices \u2013 a test inherently associated with human qualities. It is therefore unclear, in the absence of case law, how a work created entirely by a non-human could satisfy this threshold.<\/p>\n
Even if this tension were resolved, identifying the \u201cauthor\u201d of an AI-generated work is challenging. The statutory test asks who \u201cundertakes\u201d the arrangements necessary for the creation of the work. In the context of GenAI, this could theoretically be the user \u2013 but it seems doubtful that merely inputting prompts and editing outputs would satisfy this test. The language appears more naturally to describe a developer who designs and trains the underlying GenAI model.<\/p>\n
New Zealand\u2019s recognition of copyright in \u201ccomputer-generated works\u201d is relatively anomalous internationally. Copyright laws in the US, Australia, and the EU all generally require human authorship, meaning wholly AI-generated works are ineligible for protection. As for the UK (on whose legislation the New Zealand provision was modelled), the UK Government is now proposing to remove protection for works created solely by AI, while continuing to protect AI-assisted works where sufficiently original. If adopted, that amendment could influence similar reform in New Zealand.<\/p>\n
Contractual arrangements<\/p>\n
Regardless of the position under the Act, the ownership and use of GenAI outputs may also be affected by contractual arrangements. Most GenAI providers impose terms of use that govern the outputs their systems produce. These terms vary considerably but may include assignment provisions (transferring ownership to the provider or to the user), licences allowing the provider to use outputs for training or other purposes, or restrictions on how outputs may be used commercially.<\/p>\n
Businesses using GenAI tools should review the applicable terms of use carefully before relying on outputs for commercial purposes. Key questions include whether the provider claims any ownership or licence rights over outputs, whether there are restrictions on use (for example, prohibitions on using outputs in competing products), and whether the terms permit the business to assert copyright in the outputs against third parties.<\/p>\n
Infringement risk<\/p>\n
A related issue is whether using GenAI tools exposes users or developers to liability for copyright infringement. This risk may arise where a GenAI tool generates content that closely replicates material protected by a third party\u2019s copyright, or where the process of training AI models involves copying or using copyrighted works without authorisation. In New Zealand, \u201ccopying\u201d is broadly defined under the Act and includes storing or reproducing works in any material form. Unless such use is covered by a licence or a statutory exception (such as fair dealing for research or private study), training AI models on copyrighted works may constitute infringement.<\/p>\n
This issue has been tested in several high-profile cases overseas. A notable example is\u00a0Bartz et al v Anthropic PBC,\u00a0a US class action in which various authors alleged that Anthropic\u2019s AI models were trained on copyrighted books obtained from pirated \u201cshadow libraries\u201d. In June 2025, the court held that training AI on lawfully acquired books was \u201cquintessentially transformative\u201d and protected as fair use, but that downloading and retaining\u00a0pirated\u00a0copies was not. Anthropic subsequently agreed to a US$1.5 billion settlement covering past use of pirated works and requiring destruction of the infringing datasets. The case demonstrates the significant risks of using unlicensed data for AI training.<\/p>\n
Governments overseas continue to grapple with how to regulate AI training on copyrighted content. <\/p>\n
In 2025 the British government proposed allowing AI developers to use copyright-protected content for training without obtaining a licence, unless the rights holder had expressly opted out. However, in March 2026 it stepped back from this proposal, citing strong opposition from the creative industries, and is now considering introducing possible \u201cinput transparency\u201d obligations \u2013 though there are no immediate legislative proposals.\u00a0<\/p>\n
The EU has taken a stricter approach: in January 2026, the European Parliament\u2019s Committee on Legal Affairs called for mandatory transparency obligations on AI training data, a register of works that cannot be used for AI training, and a rebuttable presumption of infringement where transparency requirements are not met.<\/p>\n
In the US, a\u00a0National Policy Framework on Artificial Intelligence\u00a0(issued by the White House in March 2026) advocates a \u201cminimally burdensome\u201d approach and asserts the view that training AI models on copyrighted materials should not in itself breach US copyright law. At the same time the paper notes that the administration \u201cacknowledges arguments to the contrary exist and therefore supports allowing the courts to resolve this issue\u201d.<\/p>\n
Implications for NZ businesses<\/p>\n
While the legal framework around AI-generated content remains unsettled, New Zealand businesses using GenAI should take a pragmatic approach to managing risk and protecting their interests. <\/p>\n
Document the creative process: To support claims of originality and authorship, businesses should record the human involvement in creating AI-assisted works \u2013 particularly for content with significant proprietary value. Relevant records could include prompts, iterative drafts, and evidence of creative input and editorial judgment.<\/p>\n
Consider the \u2018computer-generated works\u2019 regime carefully: Asserting copyright in works with no human author will likely require demonstrating oversight of the \u201carrangements\u201d for their creation \u2013 for example, showing that the business developed the models or algorithms used, or that a third-party developer has assigned any relevant copyright.<\/p>\n
Consider contractual protections: Where using third party models or where contractors or agencies are engaged to create content, ensure appropriate contractual protections are in place. Where negotiable this should include clear warranties as to ownership of inputs and outputs, and transparency requirements regarding the AI\u00a0systems used.\u00a0<\/p>\n
Manage infringement risk: Mitigating the risk of copyright infringement can be challenging given that obtaining comprehensive licensing permissions will rarely be practicable (particularly where there is limited visibility over third party AI models). Businesses should therefore consider the warranties and indemnities offered by AI providers. Where they are limited, businesses may have to accept some level of risk and focus on building internal processes for vetting AI-generated content. Staying across legal developments in New Zealand and overseas \u2013 particularly in jurisdictions actively grappling with these issues \u2013 will also help businesses adapt as the law evolves.<\/p>\n
New Zealand businesses that take these proactive steps now will be best placed to capitalise on the opportunities GenAI offers. In addition, staying across legal developments in New Zealand and overseas \u2013 particularly in jurisdictions actively grappling with these issues \u2013 will also help businesses adapt as the law evolves.<\/p>\n","protected":false},"excerpt":{"rendered":"Comment: For New Zealand businesses and content creators, there is a growing need to understand how copyright law…\n","protected":false},"author":2,"featured_media":447586,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[365,228128,363,364,228129,138,11220,6712,111,43,7439,139,69,2114,113775,145],"class_list":["post-447585","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-ai-for-business","tag-artificial-intelligence","tag-artificialintelligence","tag-bell-gully","tag-business","tag-comment","tag-legal","tag-new-zealand","tag-news","tag-newsroom-pro","tag-newzealand","tag-nz","tag-opinion","tag-partner-content","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/447585","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/comments?post=447585"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/447585\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media\/447586"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media?parent=447585"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/categories?post=447585"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/tags?post=447585"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}