AI SaaS Terms And Conditions In New Zealand

Priya runs a Wellington SaaS startup that helps eCommerce brands write product descriptions. Her team has just added an AI feature that generates draft copy from a customer's product data, and the first big customer is asking a fair question: who is responsible if the AI output is wrong?

That's exactly where your terms and conditions start doing real work. AI can make your product more useful, but it also creates a few legal and commercial grey areas that ordinary SaaS terms often don't cover well.

If you're adding AI features to a New Zealand SaaS platform, your customer-facing terms should explain what the AI does, what it doesn't do, how customer data is handled, and where responsibility sits if something goes wrong. This doesn't have to be complicated, but it does need to be clear.

Do You Need Separate AI SaaS Terms And Conditions?

Usually, yes - or at least AI-specific clauses within your existing SaaS terms and conditions. Standard SaaS terms tend to cover subscriptions, access, payment, support and acceptable use, but AI features raise extra issues around outputs, data use, third-party models and reliance.

Your terms should match how your product actually works. For example, a tool that summarises meeting notes has a different risk profile from a tool that generates medical triage suggestions, financial forecasts or employment performance ratings.

For most AI SaaS products, your customer-facing terms should cover:

• what your AI feature is designed to do and any key limits on that use
• whether AI outputs are drafts, suggestions or automated decisions
• who owns inputs, prompts, customer data and outputs
• whether customer data may be used to improve, train or fine-tune models
• how personal information is handled under the Privacy Act 2020
• what users can't do with the tool
• your disclaimers around accuracy, completeness and fitness for purpose
• liability caps, exclusions and indemnities
• your reliance on third-party AI vendors or model providers

Templates can be a useful starting point, but AI clauses work best when they're tailored to your model, your customers and the practical risks in your product.

Who Is Responsible For AI Outputs?

It depends on the output, the customer, and what your terms say. Your terms should make it clear that AI outputs are generated from inputs, model behaviour and available data, and that customers are responsible for reviewing outputs before relying on them.

This is one of the most important clauses in AI SaaS terms and conditions in NZ because customers may assume that if your platform produces something, your business guarantees it. That's rarely a position you want to take.

A sensible AI output clause often says that:

• outputs may be inaccurate, incomplete, biased, outdated or unsuitable for a particular purpose
• similar or identical outputs may be generated for different users
• outputs aren't a substitute for professional advice where specialist judgement is needed
• customers must review and verify outputs before using them
• customers remain responsible for decisions made using outputs

Say your product generates draft ad copy for online retailers. If the AI suggests that a skincare product is clinically proven when it isn't, that could create risk under the Fair Trading Act 1986 if the customer publishes it. Your terms should require the customer to check claims before publication, especially where claims relate to health, pricing, performance or compliance.

That doesn't mean you can simply wash your hands of everything. If your marketing says the tool produces legally compliant product labels every time, but your terms say outputs may be wrong, that mismatch can create problems. Your website, sales materials and terms should tell a consistent story.

Can You Disclaim Liability For AI Mistakes?

Yes, you can limit some liability, but you can't exclude every risk. In New Zealand, your terms need to work alongside laws like the Fair Trading Act 1986, the Consumer Guarantees Act 1993 and contract rules around unfair or unreasonable terms.

A well-drafted liability clause can help protect your business from open-ended claims. It might cap your liability at a certain amount, exclude indirect loss, and make clear that you're not responsible for how a customer chooses to use unreviewed AI outputs.

Common liability protections include:

• a cap linked to fees paid in the previous 6 or 12 months
• exclusion of lost profits, lost revenue, lost data and reputational damage, where appropriate
• no liability for outputs that the customer has edited, published or relied on without review
• no liability for issues caused by third-party AI providers, outages or model changes outside your control
• customer indemnities for unlawful inputs, prohibited uses and published outputs

If you supply to consumers, the Consumer Guarantees Act 1993 may give customers rights that you can't simply contract out of. If you supply to business customers, you may be able to contract out of the Consumer Guarantees Act where both parties are in trade and it's fair and reasonable to do so.

The Fair Trading Act also matters. You can't use your terms to protect misleading sales claims. For standard form small trade contracts, unfair contract term rules may apply, so very one-sided liability clauses can be challenged. Small trade contracts generally capture standard form business contracts where the annual value of the trading relationship is under the statutory threshold, which is currently $250,000.

The practical point is simple: your liability clause should be firm, but not unrealistic. A clause that matches your pricing, customer type and risk profile is more likely to be useful than a sweeping disclaimer no customer believes.

How Should Your Terms Deal With Customer Data In AI Tools?

Your terms should clearly explain what customer data is collected, how it's used, and whether it's sent to third-party AI vendors. If personal information is involved, you also need to think about the Privacy Act 2020.

AI products often blur the line between product data, user prompts, files, transcripts, chat logs and analytics. Your customer needs to know what happens when they upload information into the platform.

Customer Data And Training

One of the biggest questions is whether customer data will be used to train or improve AI models. If you don't need to use customer data for training, say so clearly. If you do want that right, your terms and privacy materials should explain it in plain English.

Your clauses may need to cover:

• whether prompts, uploads and outputs are used to improve your service
• whether data is used to train a general model, fine-tune a private model, or only provide the service
• whether customers can opt out of training uses
• whether data is de-identified or aggregated before use
• how long prompts, logs and outputs are retained

Be careful with language like we never access customer data if your support team can view logs or your AI vendor processes prompts. Overpromising on privacy is a common way SaaS companies create avoidable Fair Trading Act risk.

Personal Information And The Privacy Act 2020

If your AI feature handles personal information, your business will usually have obligations under the Privacy Act 2020. This can include customer names, email addresses, voice recordings, chat transcripts, employment information, health information or any other information about an identifiable person.

In practical terms, you should consider:

• what personal information you collect through the AI feature
• whether the collection is necessary for your lawful purposes
• what you tell users at or before collection
• who you disclose personal information to, including overseas AI vendors
• how you keep personal information secure
• how people can access and correct their personal information
• what happens if there's a privacy breach

New Zealand's Privacy Act includes rules around disclosing personal information overseas, including Information Privacy Principle 12. If customer data is sent to an overseas model provider, cloud host or analytics tool, your privacy documents should reflect that.

If your customers are based overseas, you may also need to consider other privacy regimes. For example, a New Zealand SaaS business serving Australian or EU customers may have extra requirements depending on how the product is sold and used.

What Should Your Acceptable Use Rules Say About AI?

Your acceptable use rules should tell customers what they can't do with your AI features. This protects your platform, your other customers and your relationships with AI vendors.

AI tools can be misused in ways ordinary SaaS platforms might not be. Your terms should prohibit uses that are unlawful, harmful, deceptive or outside the purpose of the service.

Your acceptable use clause might ban customers from using the AI feature to:

• create misleading, deceptive or unlawful content
• infringe copyright, trade marks or confidential information
• generate spam, phishing content, malware or harmful code
• harass, discriminate against or profile people unlawfully
• make high-risk decisions without appropriate human review
• upload personal information without the right permissions
• reverse engineer, scrape, benchmark or misuse the model or platform
• avoid usage limits, security controls or subscription restrictions

Many AI vendor terms also require you to pass certain restrictions through to your own customers. If your platform relies on a third-party model provider, your customer terms should line up with your vendor terms. Otherwise, you may promise customers something you're not allowed to provide.

For example, if your AI vendor bans using the model for biometric identification, regulated medical advice or automated employment decisions, your terms shouldn't allow customers to use your product for those purposes.

Do You Need Human Review Clauses?

Yes, if customers could rely on AI outputs to make decisions, publish content or affect people. A human review clause helps set expectations that your tool supports decision-making rather than replacing human judgement.

This is especially important where outputs could affect:

• health, safety or wellbeing
• employment, recruitment or performance management
• financial decisions, credit or insurance
• legal rights or compliance obligations
• children, vulnerable people or sensitive information
• public statements, ads or product claims

Imagine your SaaS product helps HR teams screen job applications. If the AI ranks candidates, your terms should make it clear that the customer must review recommendations before making decisions. You may also need to tell customers not to use the tool in a way that breaches employment, privacy, human rights or anti-discrimination laws.

If your own team uses contractors to label data, test prompts or review outputs, your internal contracts matter too. From 2026, New Zealand's Employment Relations Act 2000 changes include a contractor gateway test, which is designed to give more certainty around contractor status where the gateway is met. If your AI operations rely heavily on contractors, it's worth checking your contractor arrangements are set up properly.

How Do Third-Party AI Vendor Terms Affect Your Customers?

Third-party AI vendor terms can directly affect what you can promise your customers. If your SaaS product is built on an external model, your customer terms need to reflect that dependency.

Your AI vendor may control key parts of the service, such as:

• model availability and uptime
• rate limits and usage restrictions
• content moderation rules
• data retention and training settings
• where data is processed
• changes to model behaviour, pricing or features
• categories of prohibited use

This matters because your customer may see your platform as one service, even if a third-party model sits behind it. If the model provider changes its API, withdraws a model or suffers an outage, your terms should give you enough flexibility to manage that without breaching your own contract.

Useful clauses can include rights to:

• modify, suspend or replace AI features where a vendor changes its service
• apply usage limits or fair use rules
• remove outputs or block prompts that breach policies
• change supported models or providers
• pass on pricing changes where your subscription model allows it
• suspend accounts where continued use would breach vendor terms or law

You should also check whether your vendor terms give you enough rights to use the model in a commercial product. Some tools have different rules for free, testing, enterprise and production use.

What Other Documents Should Sit Around Your AI SaaS Terms?

Your customer-facing terms are the starting point, but they're rarely the only document you need. AI SaaS products usually need a small set of documents that work together.

Depending on your product, you may need:

• SaaS terms and conditions: the main contract covering access, fees, accounts, IP, AI outputs, liability and termination
• Privacy policy: a public-facing document explaining how you collect, use, disclose and store personal information
• Data processing terms: useful for business customers who need more detail about security, subprocessors and overseas transfers
• Acceptable use policy: a focused set of rules for customer behaviour, prohibited content and misuse
• AI product disclaimers: short, visible notices in the product where users see outputs
• Vendor and subcontractor agreements: contracts with model providers, developers, data labellers and support providers
• IP and confidentiality agreements: especially if contractors or collaborators help build your model, prompts, datasets or code

For New Zealand companies, it's also worth keeping your corporate basics tidy with the Companies Office, your NZBN details and clear records of who owns the product IP. If you plan to protect your brand, trade mark applications go through IPONZ.

Strong AI terms are easier to draft when your product team, sales team and legal documents all say the same thing. If sales promises one level of accuracy, your help centre says another, and your terms disclaim everything, customers can lose trust quickly.

Key Takeaways

• AI SaaS terms should deal specifically with outputs, customer data, acceptable use, human review and third-party model dependencies.
• Your terms can limit liability for AI mistakes, but they still need to work with New Zealand laws like the Fair Trading Act 1986, Consumer Guarantees Act 1993 and Privacy Act 2020.
• Be clear about whether customer data is used for training, improvement, analytics or only to provide the service.
• AI output clauses should set expectations around accuracy, uniqueness, professional advice and customer review.
• If you rely on third-party AI vendors, your customer terms should reflect their limits, restrictions and service dependencies.
• Your SaaS terms, privacy policy, acceptable use rules and vendor contracts should all fit together.

If you'd like help with your AI SaaS terms, customer data clauses, acceptable use rules or vendor risk allocation, you can reach us at 0800 002 184 or team@sprintlaw.co.nz for a free, no-obligations chat.