Can You Trust the Machine? What New Zealand's Visa Software Failures Teach Us About Automation in Hiring and Migration

Meta description: Immigration New Zealand's ADEPT platform approved dozens of applications with no human review. Here is what the episode reveals about automated decision-making, and why the lessons reach far beyond visas.

Automation promises speed, consistency and lower costs. It also raises a question that every organisation adopting it eventually has to answer: what happens when the system does something nobody designed it to do? New Zealand has just given a live example. According to RNZ on 30 September 2026, briefings to Immigration Minister Erica Stanford revealed that faults in the government's online visa platform approved 32 visitor visas and 23 Job Check applications without any officer assessing them.

What went wrong, in plain terms

The platform is ADEPT, launched by Immigration New Zealand in 2022. Officials said a technical fault sometimes allowed the same automated process to run more than once on a single application. That duplication triggered repeated actions and bypassed the manual assessment step entirely.

Nobody chose this outcome. The agency had reportedly not yet used its legal power to automate decisions, and in 2025 said it was around two years from having the data, testing and quality assurance to do it well. The system effectively did on its own what the organisation wasn't ready to let it do.

A pattern, not a one-off

The auto-approvals were not the only fault reported:

• A screening question that never appeared: In 2025, 364 applicants for multiple entry visitor visas for parents and grandparents weren't asked about time spent in a high tuberculosis country. Of these, 295 were granted visas.
• Employer records that vanished: A glitch affected around 2,300 businesses whose accreditation had lapsed and been renewed, removing their migrant workers from dashboards.
• A submission bug after an update: In August 2026, some visitor visa applicants couldn't submit drafts once a second person was added.

Each is a different failure, but they share a root: a complex system changing over time, with gaps in testing and monitoring that only surfaced after the fact.

Five lessons for any organisation using automation

1. Automation can fail silently. The visa approvals were found when staff noticed one case, not because an alarm went off. Build monitoring that looks for unexpected outcomes, not just error messages.

2. Human checkpoints only work if they can't be skipped. A manual review step is a control, and controls need their own safeguards against being bypassed.

3. Audit the past, not just the present. The sweep found approvals stretching back to 2022. If you deploy a tool that screens candidates, scores applications or approves requests, periodically review its historical outputs.

4. Fixes take time, so plan for interim risk. Here the correction took two months, partly complicated by wider Microsoft issues. Have a fallback process ready.

5. Be honest about scale and impact. Officials reviewed all 32 visitor visas, reopened six for full assessment and recommended one for decline. Clear, specific accounting builds more trust than vague reassurance.

The cost of the system itself

The technology programme behind ADEPT, Our Future Services, was budgeted at $336 million. Ministers sent it back for a new business case after its projected savings were overstated by $134 million, and they want a report before April 2027. Reporting also says roughly $25 million was spent in its first year. It is a reminder that the business case for automation deserves the same scrutiny as the code.

Relevance for HR, recruitment and talent teams

Many hiring teams now use software that filters CVs, ranks candidates, schedules interviews or tracks right-to-work checks. The same questions apply:

• Who reviews the tool's decisions, and how often?
• Can a step be skipped without anyone knowing?
• Do you keep independent records that don't depend on the platform?
• What is your process when the vendor announces a fault?

For international hiring, where visa compliance and job checks are involved, a platform error can turn into a real compliance or candidate-experience problem. Keep your own evidence, and don't treat any single system as the only source of truth.

The takeaway

The debate over whether governments and employers should automate approvals will continue. What this episode shows is that the bigger risk isn't automation by design, but automation by accident. Whatever you adopt, test it thoroughly, monitor it continuously, and keep a human accountable for every decision that matters.

Building a hiring process that combines smart technology with sound oversight? Explore more insights from TALYNX.

This article is general information, not legal or immigration advice. Check official sources or consult a qualified adviser about your circumstances.