What the Commission published

On 18 November 2026 the Commerce Commission released guidance on how the Fair Trading Act 1986 and the Commerce Act 1986 apply when businesses use AI. The package sits in the Commission's business obligations section. It has a page on each Act and a longer page on algorithmic collusion (AI guidance for businesses). No new law comes with it. The guidance explains how the regulator will read existing law when a chatbot, a pricing model or an autonomous agent stands between a firm and its customers.

Deputy Chair Anne Callinan stated the principle in one sentence: "Whether a decision is made by a person, an algorithm, or an AI system, the business remains accountable for the outcome" (Commerce Commission). In the same announcement Callinan said policymakers may later consider whether extra regulatory tools are needed. For now, the Commission wants boards to treat AI as a governance issue and to have visibility over where AI is used across the organisation.

For people who build AI products, the guidance matters because it turns model behaviour into legal exposure for whoever deploys the model. A hallucinated refund policy, an invented product capability or a pricing recommendation trained on a rival's data now has a named regulator and a published test.

Consumer law: when a model writes the claim

The Fair Trading Act page says the same rules apply whether information comes from a person, an algorithm, an AI tool or an AI agent. The Commission names several risk areas. Misleading content covers AI-generated ads, website copy, images, recommendations and chatbot replies. Claims to consumers need reasonable grounds, and the page warns that AI output doesn't prove a claim is true. Synthetic images, video or audio that falsely suggest someone endorsed a product count as fake endorsements, and AI-written reviews that don't reflect real customer experiences count as fake reviews. The page also names "ghost shops", which are AI-built websites posing as New Zealand retailers when no legitimate business exists.

Two of the risk areas apply directly to product teams. Personalised pricing and offers must be clear and accurate, and the Commission says a business may need to disclose its use of AI to avoid giving a misleading impression. Persuasive design, including personalised prompts that pressure a consumer, must stay within the law on misleading, deceptive and unfair conduct. The guidance also addresses "AI washing", the practice of overstating what a product's AI does. Labels such as "AI-powered" or "AI-enabled" have to be accurate.

Penalties under the Act reach $200,000 for an individual and $600,000 for a business, and each breach can attract its own penalty. The per-breach rule matters for AI systems. In its analysis of the guidance, Bell Gully notes that a systemic error in a consumer-facing tool can reach many customers at once. A human support agent who gets a refund policy wrong affects one conversation. A chatbot with the same error repeats the mistake in every session until someone notices.

Competition law: pricing engines and shared tools

The algorithmic collusion page confirms that the cartel provisions in section 30 of the Commerce Act apply whether people, software or AI make the decisions. The Commission describes several ways collusion can happen through algorithms. Competitors can use software to carry out or monitor a price agreement, and the page says that automating the behaviour doesn't make it lawful. Algorithms can be fed rivals' current or future pricing, customer or strategic data. Competitors can also share a common pricing platform, data intermediary or AI service that pools their confidential information. The page warns that relying on an independent provider doesn't automatically protect a business. It also says research suggests some advanced systems with limited human oversight may learn behaviour that produces coordinated market outcomes.

The Commerce Act page covers misuse of market power. Its examples include a platform using ranking or recommendation systems to favour its own products, a cloud provider requiring customers to take its AI products as part of a wider package, and AI-driven pricing set to exclude competitors. The page also says merger law can apply to acquisitions of technology, data assets and important AI capabilities, and it names "killer acquisitions", where a business buys an innovative start-up to remove future competition. Founders talking to larger acquirers should expect that deal to be assessed against this framing.

What builders can do with the guidance

The Commission's recommended steps are written for compliance teams, but most of them map onto engineering work. Our reading of the official pages suggests these concrete tasks:

  1. Keep an inventory of every customer-facing model, including AI embedded in third-party platforms, with a named owner for each. Bell Gully's summary lists mapping and assigning responsibility among the recommended steps.
  2. Build an evaluation set of realistic customer questions about refunds, guarantees and product capabilities, then run it against every prompt or model change. The Commission asks businesses to regularly test customer-facing tools, including chatbots.
  3. Store the evidence behind product claims, including any claim about what your AI does, so a "reasonable grounds" question can be answered from records.
  4. For pricing products, document data sources and keep each customer's data separate. The collusion page tells buyers to ask vendors whether competitor information feeds outputs and whether data is separated between businesses, so vendors should have the answers written down.
  5. Design override paths so staff can understand, challenge and reverse a model's recommendation. Log design decisions and audit results, because the Commission says good records can help show that decisions were made independently.

A New Zealand company selling one pricing or recommendation tool to several competing retailers is the clearest case where the guidance becomes a sales question. Customers following the Commission's vetting advice will ask for the same documentation, and a vendor with tenant isolation and audit logs already in place will answer faster.

Limits and open questions

The guidance has gaps. The pages contain no case studies or worked scenarios, and the collusion page gives no penalty amounts beyond saying penalties can be substantial, with criminal liability possible for individuals in some cases. The Commission doesn't define how much oversight counts as "meaningful", how often testing should run, or what disclosure wording would satisfy the personalised pricing advice. Every page states that the material isn't definitive and doesn't replace legal advice. The guidance only covers the two Acts the Commission enforces, so it says nothing about other obligations an AI product may carry.

Builders who want to put those questions to the regulator have an opportunity next week. The AI Forum is hosting a free Zoom webinar with the Commission on 24 November 2026 from 12pm to 1pm NZ time, with Anne Callinan and Will Costello, the Commission's Principal Adviser Competition. Members who want to talk through how the guidance affects their own products can also find society meetups on our events page.

Sources

Build with people working on the same things

Join the Auckland Society for AI: NZ$599 a year with a ForgeBox and the members' projects area. Our Auckland meetups are open to everyone.

Become a member