AI, Law, and Policy in Aotearoa

Overview

The uptake of AI is pushing the boundaries of New Zealand's law and of how our government works. Generative AI tools built into social media platforms are producing harms, such as nonconsensual sexualised deepfakes, that the Harmful Digital Communications Act 2015 and common law were never designed for. Government agencies are adopting AI faster than the guidance they work from can keep up, and the national AI strategy released in July 2025 is built around economic opportunity with little on managing ethical and social risk. And courts hold decades of structured decisions that machine learning can now predict, which raises the question of whether a predicted sentence can be explained well enough to be trusted at all. Working with legal scholars and policy researchers, this project studies all three angles: whether the law as written can address AI-enabled harms, what a small state should actually do about advanced AI, and whether models trained on legal data can be made to show their reasoning.

Themes

• Can the law address AI-enabled harms? With Dr Cassandra Mudgway (Faculty of Law, University of Canterbury) and Chris McGavin (LensenMcGavin AI), we investigated whether New Zealand's legal framework can respond to nonconsensual sexualised deepfakes produced by generative AI built into social media platforms. Using the Grok deepfake incident on X as a case study, and taking a feminist legal and technofeminist lens, we evaluated the Harmful Digital Communications Act 2015, the privacy torts, defamation, and negligence. Each has limited meaningful application to these harms. The proposed Deepfake Digital Harm and Exploitation Bill would close some definitional gaps, but criminalisation alone cannot address the systemic risks created by platform design, business incentives, and inadequate safeguards. The article, Grokking Misogyny, is forthcoming in the New Zealand Universities Law Review. This work underpins our joint call for risk-based AI regulation across all areas of law.
• Re-engineering government guidance. Johniel Bocacao's MSc thesis, Unified Guidance Is All You Need, co-supervised with Prof. Ali Knott, mapped the guidance New Zealand government agencies have on trustworthy AI and showed where it overlaps, conflicts, and leaves gaps. He proposed a New Zealand Artificial Intelligence Manual (NZAIM), modelled on the government's information security manual, to unify that guidance into something an agency can follow from procurement through to deployment.
• The policy and practice of AI security. Ben Cravens's MSc thesis, The Policy and Practice of AI Security, also with Prof. Ali Knott, works in two halves. The first develops scenarios for advanced artificial intelligence and works through the risks and policy responses each implies for New Zealand: a structured basis for planning that does not depend on any one forecast of how capable the technology becomes, or how fast. The second turns from policy to practice, and asks how large language models can actually be aligned with the values and rules we set for them.
• Explainable sentence prediction. With Dr Marcin Betkier (Faculty of Law), Harry Rodger built a proof-of-concept explainable model that predicts imprisonment sentences for assault from the text of New Zealand judgments. Its predictions were accurate to within one year, and because the model is explainable we could see which phrases in a judgment drove each prediction and weigh the benefits and risks of using such a model in the courts. The work was published in the Journal of the Royal Society of New Zealand and discussed in The Conversation and on RNZ's The Panel.

The public advocacy that grows out of this work, including the open letter to Parliament calling for risk-based regulation, is on the engagement page.

Team and collaborators

Selected outputs

• Mudgway, C., Lensen, A., & McGavin, C. (forthcoming). Grokking Misogyny: Is New Zealand Law Ready for Sexualised Deepfakes? New Zealand Universities Law Review.
• Bocacao, J., Knott, A., & Lensen, A. (2026). Re-Engineering New Zealand Government Guidance for Trustworthy AI System Delivery: A Whitepaper. Te Herenga Waka—Victoria University of Wellington, 16 March 2026. Whitepaper · Project site
• Mudgway, C., & Lensen, A. (2026). Sexualised deepfakes on X are a sign of things to come. NZ law is already way behind. The Conversation
• Rodger, H., Lensen, A., & Betkier, M. (2022). Explainable Artificial Intelligence for Assault Sentence Prediction in New Zealand. Journal of the Royal Society of New Zealand. PDF

See the publications and media pages for more.