November 23, 2026

If AI wrote most of your code, your IP schedule may be fiction

A developer writes code on a laptop in front of multiple monitors in an office setting.

The argument about whether AI can be creative is a distraction. The question that matters for business is narrower and more expensive: when an asset was built with generative AI, can you sell it, licence it, warrant it and sue over it? Right now, for a lot of New Zealand firms, the honest answer is “probably, but we couldn’t prove it.”

That uncertainty sits inside marketing retainers, software licences, publishing deals and design contracts. And the rules are about to be rewritten.

The law already has an answer, sort of

New Zealand is not starting from zero. Under section 5(2)(a) of the Copyright Act 1994, a work made with a generative AI tool can potentially be protected, with the person who input the parameters or made the arrangements deemed the author and owner. That is more generous than the United States, where the Copyright Office has refused to register AI-generated works even with extensive prompting.

But ownership on paper is not the same as a defensible asset. Originality still turns on “the degree of human skill, labour, and judgement” applied, according to Bell Gully. The less a human did, the weaker the claim. The firm’s practical advice is blunt: record the creative process and the human input so you can support originality later. Very few agencies or dev teams do this systematically.

There is also no standard international approach, meaning NZ exporters face inconsistent treatment the moment their AI-assisted work crosses a border.

Your vendor may own more than you think

The under-discussed risk is contractual. Bell Gully notes that ownership and use of AI outputs may also be affected by the terms of the AI provider itself, including assignment clauses, licences letting the vendor reuse outputs for training, or restrictions on commercial use. A business can own an asset by statutory default while its terms of service say something different. Most procurement teams have never read those terms against a client contract that promises clean title.

Software has the most to lose

Visiting Stanford scholar Mark Lemley, described as the world’s most-cited IP academic, put it plainly last month: “You’ve got a real problem, if you’re a software company”, because code is now overwhelmingly written with AI. He cites 70% of new code as AI-generated, and points to more than 150 pending lawsuits worldwide testing both training and copyrightability.

If large slices of a codebase are held to be uncopyrightable, the proprietary moat shrinks. For a SaaS company, that hits enterprise value directly. Expect acquirers and investors to start asking in due diligence what share of the code was AI-generated, what human review it received, and whether that is documented. Founders who cannot answer will take a haircut on price or warranties.

Two deadlines, one review

This is no longer academic. Cabinet has asked the Minister of Commerce and Consumer Affairs to report back by 31 March 2027 on a copyright framework for generative AI. Separately, New Zealand has until mid-2028 to update copyright law under the EU free trade agreement. In March 2026, Minister Scott Simpson acknowledged how far technology had moved since the last major review in 2008.

The issue has been on officials’ radar for a while. Back in 2024, Treasury flagged that copyright law might need updating for AI and suggested aligning with trading partners where appropriate.

The academic framing is taking shape too. Auckland Law School’s Joshua Yuvaraj argues protection should rest on genuine human creative input, but does not exclude AI-assisted work: “Where a person contributes genuine skill and labour, including through instructions given to an AI system, copyright protection may still be justified.” Earlier this year he floated a reduced 50-year term for AI-assisted works, building on his argument that AI is not fully creative. A shorter term is workable for business. What matters is that it is clear.

A licensing bill is on the table

The creative lobby is pushing harder. Its latest policy paper says the sector supports 118,000 jobs across 37,504 businesses and $17.5 billion of GDP, and proposes a licensing-first regime requiring any organisation doing business in NZ to license copyrighted works used to train, prompt or generate outputs from AI. Creators have a legitimate grievance; 99% of surveyed creators say their work was scraped without consent. But as drafted, that is a compliance obligation reaching well beyond the arts, into any firm fine-tuning a model on third-party content. Ministers should get the detail right rather than bolt on a broad new licensing bureaucracy.

What boards should do before March

Don’t wait for the report-back. Audit which client-facing assets and products are AI-assisted. Read your AI vendors’ terms against the IP warranties you give customers. Start logging human input as a matter of routine. And if you plan to sell or raise capital in the next two years, assume someone will ask how much of what you own a machine actually made. The firms that can answer that question cleanly will be the ones whose IP still counts as an asset when the new rules land.

Sources

Reader Poll · 6 questions

Do you agree or disagree with the following?

Community

Join the discussion

Add useful context, ask a good question, or challenge an idea — keep it specific and respectful.

Create a commenter account

Enter the name you want shown publicly and your email. We will email you a password-set link; you cannot comment until you use it.

Your email is used for sign-in and account security. It is not published with comments.

Subscribe for weekly news

Subscribe For Weekly News

* indicates required