The Commerce Commission has released its regulatory interpretation on the commercial deployment of artificial intelligence. Issued on 18 November 2026, the guidance for New Zealand businesses using AI clarifies how existing statutes apply to algorithmic tools. Crucially, this intervention does not represent a new AI law, nor does it establish new legal precedent or automatic, exclusive strict liability for every technological misstep. Instead, it serves as a restatement of existing obligations under the Fair Trading Act 1986 and Commerce Act 1986, viewed through the lens of emerging technology.
For policy analysts and corporate governance teams, the framework prompts a central, operational question: what specific evidence would satisfy the regulator that a business exercised appropriate control over a particular AI use? By separating foundational legal obligations from suggested technological controls, the Commission has outlined a risk-based approach to market integrity.
Proportionality and the Fair Trading Act
The guidance encompasses a broad spectrum of artificial intelligence applications, which necessitates a nuanced reading of proportionality. From a risk management perspective, using generative text to draft internal marketing copy carries a vastly different operational profile to deploying a customer-facing chatbot, which in turn differs substantially from authorising an action-taking autonomous agent to execute dynamic pricing. The Fair Trading Act component of the guidance outlines risks such as AI-generated unsubstantiated claims, fake reviews, misleading digital storefronts, and manipulative marketing.
However, the suggested controls highlight a distinct supplier information imbalance. Small and medium enterprises (SMEs) procuring third-party AI often lack visibility into how these models are trained or weighted. While the regulator suggests businesses review AI-generated content and test customer-facing tools, practical compliance requires proactive procurement evidence. We recommend businesses require software vendors to provide explicit documentation detailing their data sources and algorithmic guardrails. This documentation acts as vital evidence that the deploying business took reasonable steps to verify the system's accuracy before exposing it to consumers.
Furthermore, the guidance touches upon the need to disclose the use of AI in personalised pricing to avoid misleading consumers, but stops short of providing practical disclosure examples. Effective disclosure mechanisms remain a grey area for user experience designers striving to balance legal compliance with seamless commerce.
Algorithmic Collusion and the Commerce Act
The most legally intricate section of the release addresses market competition. The regulator explicitly notes that automating anti-competitive behaviour does not make it lawful. However, the detailed algorithmic collusion fact sheet provides a vital legal distinction: autonomous parallel pricing does not automatically prove a cartel. The guidance fairly credits the statutory requirement under Section 30 of the Commerce Act, acknowledging that a breach still fundamentally requires a contract, arrangement, or understanding between competitors.
This creates legal uncertainty regarding emergent coordination. If advanced, isolated AI systems independently learn to converge on supra-competitive prices using only public market data—without explicit human instruction or the sharing of sensitive commercial information—liability remains difficult to attribute. The Commerce Act web guidance rightly warns that using a shared software platform that pools competitors' confidential data can facilitate an unlawful arrangement. However, emergent, self-learned coordination by genuinely independent agents remains a complex frontier that will likely require future judicial testing.
Operational Oversight and Incident Response
To mitigate these risks, the Commission recommends maintaining human oversight over algorithmic systems. From an operational standpoint, it is important to acknowledge that human oversight cannot realistically prevent every single breach, particularly within high-frequency algorithmic environments. The regulator's fact sheet addresses this by calling for meaningful review and challenge, rather than a superficial rubber-stamping of automated outputs.
Addressing a gap in the guidance regarding remedy and incident response, we recommend that organisations implement an explicit operational "stop authority." This means establishing a clear protocol—and designated personnel—with the authority to immediately suspend or override an algorithmic agent the moment unexpected or potentially anti-competitive behaviour is detected.
Scope and Future Governance
It is worth noting that the Commerce Commission's guidance deliberately excludes matters of privacy, data protection, and employment law. Given the regulator's specific economic and consumer protection remit, this absence is appropriate and should not be viewed as a regulatory failure. Businesses must integrate this guidance with distinct advice from the Privacy Commissioner and employment authorities.
Ultimately, the controls proposed by the Commission—such as regular system audits, independent commercial decision-making, and robust compliance documentation—should be interpreted as recommended best practices for corporate governance rather than new mandatory legal requirements. They provide a roadmap for businesses to build an evidence trail, ensuring that when an algorithm inevitably falters, the organisation can demonstrate it maintained meaningful, proportionate control.
Sources
- Commerce Commission: Commerce Commission issues AI guidance for New Zealand businesses
- Commerce Commission: AI and the Fair Trading Act
- Commerce Commission: Algorithmic Collusion and the Commerce Act
- Commerce Commission: Fact sheet – Algorithmic Collusion and the Commerce Act (PDF)
Analysis. Revised to examine the guidance and its practical implications. Automated systems assisted with research and drafting; Gemini independently reviewed this revision. Primary sources are linked for verification.
