The Employment Relations Amendment Act 2026 received Royal Assent on 20 February 2026. The commentary has been accurate about its significance: this is the most consequential change to New Zealand employment law since 2018. It rewrites who qualifies as an employee, removes dismissal protections for high earners, restructures how personal grievance remedies are assessed, and scraps the 30-day rule.
It does not mention artificial intelligence once.
Not in the Act. Not in the explanatory note. Not in the select committee report. AI — the technology actively reshaping how NZ employers hire, manage performance, make redundancy decisions, and engage contractors — is entirely absent from legislation that governs all of those relationships.
This is not an oversight in the technical sense. The drafters knew what they were drafting. The select committee heard extensive submissions. The Minister has been articulate about the reforms throughout. The absence of AI is not a mistake. It is a gap, and it is a significant one, because the reforms Parliament has just passed intersect with AI in ways that will generate legal uncertainty for years.
The gateway test and AI-managed work
The Amendment's most structurally important reform is the new "specified contractor" category. Under the old framework, whether a worker was an employee depended on the real nature of the relationship — courts could look through the contractual label to the substance. The Amendment changes this: if a written agreement labels the person a contractor, they are not restricted from working for others, and they have flexibility around working hours or can subcontract, they fall into the "specified contractor" category and outside the employment framework.
This test was designed with a traditional contractor in mind — a tradesperson who sets their own hours, accepts or declines jobs as they choose, and controls their own workflow. It was not designed for a worker whose task allocation, availability incentives, performance ratings, and effective income are managed by an algorithm.
AI-managed platforms exercise comprehensive control through mechanisms the gateway test does not measure. Algorithmic task queues, dynamic pricing, acceptance rate monitoring, rating systems that penalise non-engagement — these tools shape when people work, how much they earn, and whether they receive future work, without technically restricting them from other employment or requiring specific hours.
The Act does contain one partial safeguard: s6(8) provides that if the hours of work with the principal are such that they effectively restrict the person's ability to work for others, the gateway is not met. This is precisely where AI-managed platform work will be contested. A worker can be free in the formal legal sense — no contractual restriction, no required hours — while being practically dependent on a single platform's algorithm for income. Whether algorithmic incentive structures and rating-based earnings dependency constitute effective restriction under s6(8) is a question the Act creates but does not answer.
The ERA will be asked to apply this statutory framework to arrangements its drafters did not have in mind. Does algorithmic direction constitute restriction? Does earnings dependency created by a rating system amount to control? The Act does not say, because it did not ask.
The $200,000 threshold and the good faith gap
The Amendment introduces a $200,000 total remuneration threshold — salary plus bonuses, commissions, and other components — above which employees lose the right to bring a personal grievance for unjustified dismissal. The mechanics are in s113A, and s67I goes further still: when a high-income employee is dismissed, the employer is not required to follow the good faith consultation obligations in s4(1A)(c) of the Employment Relations Act 2000, and is not required to provide written reasons for dismissal on request under s120. The procedural protections that applied to senior employees — consultation before dismissal, reasons on request — are both gone.
The employer rationale for this reform is coherent: senior roles require organisational confidence, and excessive procedural protection at the top of an organisation makes restructuring unwieldy. That is a legitimate policy argument. But it was made without considering who is most subject to AI-assisted employment decisions.
High earners are not sheltered from algorithmic management. Quite the opposite. AI performance analytics, productivity tracking, cost-value modelling in restructuring proposals, and AI-assisted benchmarking are concentrated in professional and managerial roles — exactly the employees the Amendment has just exempted from procedural protection. These are the employees whose redundancy is most likely to have been informed by an AI analysis, whose performance decline is most likely to have been flagged by an algorithm, and who now have the least recourse when those assessments go against them.
The employees most likely to be subject to AI-informed decisions are now the employees with the least procedural protection when those decisions result in dismissal.
There is no requirement to disclose AI involvement in a restructuring recommendation. There is no good faith consultation process through which a senior employee could question the methodology. There is no obligation to even provide written reasons. The threshold removes all of it.
Contributory conduct and the algorithm problem
The Amendment also significantly strengthens the contributory conduct framework. New s123B provides that where an employee's conduct contributed to the situation giving rise to a personal grievance and that conduct amounts to serious misconduct, no remedies are available at all. New s123C bars reinstatement and compensation where contributing behaviour is established — regardless of whether it reaches the serious misconduct threshold. And amended s124 allows available remedies to be reduced by up to 100%.
This framework makes straightforward sense where the employer's assessment of employee conduct is entirely human. It becomes considerably more complex where that assessment is AI-assisted. If an AI performance management system flagged declining output metrics, those metrics triggered a performance management process, and the employee's response to that process contributed to a subsequent dismissal — the contributory conduct assessment requires asking what the employee did. It should also ask whether the employee had a meaningful opportunity to understand and respond to what the AI was measuring, how it was measuring it, and whether the measurement was accurate.
The Act does not require that question to be asked. The concept of "serious misconduct" — on which the harshest remedy bar now turns — remains undefined in the amended Act. Where an employer relies on AI-generated performance data to characterise conduct as serious, and that characterisation eliminates an employee's remedy entirely, we are in genuinely uncharted territory. The ERA will define the boundaries through litigation.
The democratic deficit
There is a version of this story in which AI's absence from the Amendment is an understandable lag. Technology moves faster than Parliament, and the legislative calendar does not wait for emerging risk to fully crystallise. That version is partly true. It would be a sufficient explanation if the gap were narrow.
It is not narrow. The contractor/employee boundary is the defining employment dispute of the platform economy — and the platform economy is AI-managed. The good faith obligation being removed for high earners is the procedural safeguard against opaque decision-making — and AI is the most opaque decision-making tool most employers have ever used. The contributory conduct framework that can now eliminate remedies was written for human misconduct in a human-assessed process, not for AI-flagged performance in an algorithmically mediated workplace.
Parliament debated these reforms through the second half of 2025. The select committee heard from employers, unions, law firms, and industry groups. The Employment Relations Authority has already published guidelines on AI use in proceedings. The technology's presence in NZ employment relationships is not a future consideration — it is a present reality. None of that found its way into the Amendment.
The legislation that will govern NZ employment relationships for the next several years was written as if AI does not exist. That is a democratic deficit — not in the sense of bad faith, but in the sense that the law and the reality it purports to govern are already misaligned, and the Amendment has made that misalignment wider.
What the existing framework still requires
The Amendment does not dissolve the legal obligations that already apply to AI in the workplace. The good faith framework in s4 still requires employers to be active and communicative in employment relationships — including, for most employees, transparency about how AI-assisted decisions are made. The Privacy Act 2020 information privacy principles govern AI-processed employee data. The Human Rights Act 1993 prohibits discriminatory employment decisions, including those where a biased algorithm contributed to the outcome. The s103A justification test — whether the decision was one a fair and reasonable employer could have made — applies without modification to decisions in which AI played a role.
These existing obligations do not fill the gap the Amendment has created. They are general frameworks applied to circumstances they were never drafted to address specifically. What the Amendment needed — and did not provide — was explicit engagement with AI in employment decisions: what disclosure is required, how an employee can understand and challenge an AI-assisted assessment, and what procedural steps apply when AI is doing work the good faith framework previously assigned to human deliberation.
That question will not wait for the next reform cycle. It will be answered by the ERA.