New Zealand Considers AI-Powered Benefit Decisions Under Proposed Law
National MP Scott Simpson introduced a bill that would allow the Ministry of Social Development (MSD) to use AI to make decisions about people’s benefits. The proposed law would amend the Social Security Act to permit “automated decision-making” for benefit eligibility, payment amounts, and compliance assessments.
RNZ reports the bill has drawn criticism from privacy advocates and legal experts who warn that automated benefit decisions could disproportionately affect vulnerable populations. Critics argue that welfare recipients are least likely to have the resources or knowledge to challenge an AI decision. MSD has said any AI system would include human oversight and appeal mechanisms.
The bill is part of a broader push by the government to use AI across the public sector, as the Ministry of Business, Innovation and Employment explores general-purpose AI adoption.
Why it matters: This is a classic AI ethics dilemma playing out in real-time. Automated welfare decisions promise efficiency (faster processing, reduced administrative costs) but risk errors that compound existing disadvantage. The proposed human oversight is the key question — is the “human in the loop” a genuine reviewer or a rubber stamp? The UK’s similar Post Office Horizon scandal — where automated decisions were treated as definitive — hangs over every government AI proposal.
European Council Drafts AI Liability Directive — Winners and Losers
The European Council is circulating a draft AI Liability Directive that would harmonise how EU member states handle civil liability claims involving AI systems. Key provisions include:
- Presumption of causation: if an AI system’s output caused harm in a “reasonably foreseeable” way, the developer is presumed liable — shifting the burden of proof from the victim to the company
- Joint liability: multiple actors in the AI supply chain (data providers, model trainers, deployers) can be held jointly liable
- Algorithmic transparency orders: courts can demand access to AI system training data, weights, and logs, with protective measures for trade secrets
Tech industry groups are pushing back hard, arguing the directive will make European AI development uninsurable and drive companies offshore. Consumer and digital rights groups say the directive doesn’t go far enough and excludes “non-physical harms” like discrimination or loss of opportunity.
Axios reports the final text is expected by September 2026, with implementation by 2028.
Why it matters: The AI liability directive is the financial teeth behind the EU AI Act. The AI Act tells you what you must do. The Liability Directive tells you what happens when you don’t. The presumption of causation is the most significant provision — it flips the assumption that AI systems are unpredictable black boxes to the assumption that harms are the developer’s responsibility unless proven otherwise. This will change how every AI company operating in Europe approaches testing, documentation, and insurance.
Leiden Declaration: AI and Mathematics — A New Research Framework
Over 100 mathematicians and computer scientists have signed the Leiden Declaration on Artificial Intelligence and Mathematics, setting out principles for how AI should and shouldn’t be used in mathematical research. The declaration was released June 2.
Key principles:
- AI should be used as a research tool, not a replacement for mathematical reasoning
- Verification of AI-generated mathematical claims must remain mandatory — “no AI conjecture without proof”
- Mathematical journals should require authors to declare AI use
- AI training data that includes copyrighted mathematical works raises unresolved ethical questions
The declaration is a response to the rapid adoption of AI in mathematics, where models have generated novel conjectures, discovered new knot invariants, and even produced proofs that humans later verified (or debunked).
Why it matters: Mathematics faces a unique challenge with AI. Unlike art or writing, where AI is seen as augmenting or threatening creative work, mathematics straddles a line between “assistive tool” and “genuine research collaborator.” The Leiden Declaration tries to draw that line: AI as amplifier of human mathematical intuition, not substitute. The “no conjecture without proof” rule is the hardest one to enforce — an AI might propose a theorem that takes years to verify.
Authors Escalate Legal Action Against NZ Government Over AI Copyright
A coalition of authors and publishers escalated legal action against the New Zealand government on June 2, challenging the use of copyright-protected works in AI training without compensation. Copyright Licensing NZ CEO Sam Irvine told RNZ the government’s push for AI adoption across the public sector creates “unprecedented risk” for creators whose work is used without permission or license.
The legal action follows a broader NZ copyright review initiated by MP Cameron Brewer, who is examining how copyright law applies in the AI era. The review covers everything from music rights (Split Enz cited as an example) to written works and visual art.
The authors’ position: the government should not be exempting itself from copyright obligations for AI training data when it’s simultaneously “pushing for much more AI in the public sector.”
Why it matters: The NZ copyright AI debate mirrors the global one — every jurisdiction is trying to figure out whether AI training on copyrighted works is “fair use” (the US approach), “text and data mining exception” (the EU approach), or something else entirely. NZ hasn’t decided yet, and the authors’ legal action forces the issue. The outcome will set the precedent for AI regulation across the Pacific — Australia is watching closely.
Uber to Cut 23% of Jobs in HR Team as AI Handles Core Functions
Uber announced it will cut 23% of positions in its human resources department, with AI-based systems taking over payroll, benefits administration, recruitment processing, and compliance monitoring. The cuts affect approximately 230 roles across Uber’s global HR organisation.
Uber CEO Dara Khosrowshahi cited “operational efficiency gains from AI deployment” as the driver, noting that the company’s HR chatbot now handles 78% of first-contact employee queries without human intervention.
Why it matters: HR is often described as a “safe” profession from automation because it involves human judgment and empathy. Uber’s cuts suggest otherwise — at least for the administrative layer of HR. This will ripple through the broader HR industry, which employs hundreds of thousands of people globally. The question is whether AI-driven HR produces better (or worse) outcomes for employees than human-driven HR.
🔍 THE BOTTOM LINE
AI ethics debates are becoming concrete: liability rules that shift burden of proof onto developers (EU), welfare decisions that risk automating disadvantage (NZ), copyright fights over training data (global), and academic integrity concerns in research (Leiden Declaration). The common thread is that the “move fast and break things” era of AI is ending — and the messy work of assigning responsibility, liability, and compensation is beginning. Every jurisdiction is answering these questions differently, which means the next 12 months will produce a fragmented global AI governance landscape.
❓ Frequently Asked Questions
Q: What does the NZ benefit AI bill actually propose? It would amend the Social Security Act to let MSD use automated decision-making for benefit eligibility, payment amounts, and compliance. Human oversight is promised but not yet specified in the bill’s text. Privacy advocates warn that welfare recipients are least able to challenge AI decisions.
Q: What’s the biggest change in the EU AI Liability Directive? The presumption of causation — if an AI system’s output causes harm in a reasonably foreseeable way, the developer is presumed liable. The victim doesn’t need to prove how the model arrived at the harmful outcome; the developer must prove it wasn’t their system’s fault. This is a major shift from existing product liability law.
Q: Will the Leiden Declaration slow down AI in mathematics? The declaration isn’t binding — it’s a set of principles signed by researchers voluntarily. Its influence depends on how many journals adopt its guidelines. The most significant impact could be on the culture of AI use in mathematics: requiring AI-use disclosure and proof verification for AI-generated claims.
Q: Could NZ follow the US “fair use” model for AI copyright? Unlikely. NZ copyright law is closer to the UK and Australian model, and the current review by MP Cameron Brewer is exploring a custom approach. The authors’ legal action escalates the timeline — a court decision could define the rules before legislation is ready.
SOURCES
- RNZ — Proposed law could see government use AI to make decisions about people’s benefits
- RNZ — Authors versus AI and the risks to government public sector push
- NZ Herald — Split Enz to benefit from Government copyright changes, Cameron Brewer AI copyright review
- European Commission — Draft AI Liability Directive (Axios coverage)
- Axios — EU prep AI liability rules
- TechCrunch — European Council AI liability talks
- Leiden Declaration on AI and Mathematics — Official text
- Stocktwits — Uber to cut 23% in HR