In 2014, a group of truck drivers in Maine secured a $5-million settlement—not over harsh working conditions or wage theft, but due to the absence of an Oxford comma in the American state’s overtime law. The statute listed exempt activities as “packing for shipment or distribution”, and the lack of a comma before ‘or’ triggered a dispute. Was distribution a separate exempt activity, or was only the act of packing (for shipment or distribution) exempt? The ambiguity was sufficient for the court to side with the drivers, highlighting how even minute syntactic choices in legislative drafting can carry substantial consequences.
The case shows how language structures shape legal meaning, especially under textualist or purposivist readings. In increasingly complex legal systems, such linguistic fragilities expose the need for greater precision in drafting.
As the world changes faster than parliaments can respond, it’s easy to see why governments are starting to explore whether artificial intelligence can help. Some are already moving beyond using it to summarise bills or streamline services, towards something far more ambitious. The UAE is taking the boldest step yet, aiming to make AI a co-legislator of sorts. It has launched an ‘AI-driven regulation’ system that will not only draft and review laws, but also predict when they need to change, using a vast database of legal and public sector data. Officials hope this will make lawmaking up to 70 percent faster. Unlike many democracies, the UAE’s political structure lets it experiment at speed, which is why it’s becoming a testbed for such innovation.
Integrating AI into legislative drafting presents several advantages. First, AI can significantly enhance efficiency by automating routine tasks such as document review and due diligence, leading to substantial time savings and productivity increases. Second, AI can improve accuracy in legal research and drafting; for instance, studies have shown that AI can achieve higher accuracy rates in identifying legal risks compared to human lawyers. Third, AI facilitates consistency in legal documents by maintaining uniform language and structure, which is crucial in legislative texts. Fourth, AI can assist in identifying and mitigating biases in legal decision-making by analysing large datasets for patterns of discrimination. Fifth, AI supports real-time updates and efficient cross-referencing in drafting, enabling lawmakers to develop more informed and organised legislation.
But before embracing such a future, it’s worth examining what might go wrong. AI models trained on legal corpora are only as good as the data they ingest. If historical laws or judgements reflect biases, say, against marginalised groups, the AI won’t correct them; it will likely replicate and entrench them. More technically, transformer-based language models may misinterpret legal syntax or collapse distinct legal doctrines into statistically similar phrasing. And when AI is used recursively, when laws written by machines inform the next generation of training data, the system risks locking in errors and blind spots over time.
There’s also the black box problem. These models, particularly large language models used in natural language processing, often produce outputs that are difficult to explain even to their creators. So when an AI suggests amending a tax clause or reclassifying a criminal offence, who takes responsibility? And on what basis? Unlike humans, these systems lack intent or moral reasoning. Yet, their recommendations can carry the weight of policy.
Worse still, AI systems often overfit to structure and formalism, generating legally correct but purposively flawed drafts that fail to reflect the evolving spirit of law. Should a constitutional right be interpreted by probability-weighted token sequences? Even the UAE’s bold experiment, using AI to suggest legal updates and model societal impacts, forces us to confront uncomfortable questions: what happens when predictive efficiency collides with normative legitimacy? In such a world, democratic oversight, legal theory, and ethical reasoning must become more, not less, central to the legislative process.
Equally critical is the epistemic authority that AI may begin to command. As these systems become more capable of summarising legal precedent, evaluating policy outcomes, and generating model legislation, there’s a risk that their outputs are treated as inherently superior, even unchallengeable, due to their computational origin. In lawmaking, such ‘automation bias’ could erode deliberative pluralism. The legislative process thrives not just on efficiency but on friction, argument and dissent that reflect a society’s moral and cultural diversity. If AI shortcuts that process in the name of rationality or optimisation, we risk losing something fundamental: the idea that laws are not merely well-drafted instructions, but collective moral choices negotiated in public view.
As things stand, AI should be seen as a precision instrument—one that enhances clarity, detects inconsistencies, and streamlines the drafting of legal language—but not as a normative guide that decides where or whether the state should intervene. That judgement requires ethical reasoning, institutional memory, and a deep understanding of political context. While AI can improve the how of legislation, it lacks the legitimacy or wisdom to decide the why or what.
(Views are personal)
Aditya Sinha
Public policy professional