How artificial intelligence is radically reshaping tax law
Generative artificial intelligence (AI) is transforming tax research, compliance and regulation faster than many professionals and policymakers expected
AI is accelerating at a pace few anticipated, and for tax professionals, legal advisers, and regulatory bodies, that momentum is reshaping how tax laws are interpreted, applied, and enforced. Since AI is no longer theoretical, experts say tax professionals must now focus on using technology to streamline complexity, enhance compliance, and reshape the modern tax system.
“AI is improving really quickly,” said Professor Benjamin Alarie, Faculty of Law at the University of Toronto, speaking at the 16th International ATAX Tax Administration Conference hosted by UNSW Business School. “But I think it's getting better faster than most of us really appreciate.”
Prof. Alarie is a global expert on the intersection of law and AI technology and author of The Legal Singularity, which explores how AI systems, algorithms, and predictive models could radically improve the legal system. He is also co-founder and CEO of Blue J, a legal tech company using machine learning and generative AI to assist tax professionals, improve tax compliance, and clarify complex tax codes for taxpayers across the US, Canada, and the UK.
A turning point for AI – and for tax law
To illustrate the accelerating pace of change, Prof. Alarie referenced data insights from Metaculus – a forecasting platform that aggregates predictions on AI development timelines and scientific innovation.
Back in March 2020, Metaculus estimated that Artificial General Intelligence (AGI) – AI capable of human-level decision-making – wouldn’t arrive until 2050. But by early 2025, that forecast has leapt forward to 2026, accelerated by approximately 23 to 24 years due to breakthroughs in language models, generative AI, and applied AI algorithms.
Prof. Alarie also talked about AI’s specific robotic capabilities – like the ability for AI to build a 1:18 scale model of a 2021 Ferrari 312 T4. The model is complex, with hundreds of parts, and the instruction manual is 452 pages long. ChatGPT estimated it would take a human 40 to 80 hours to complete. So, when will an AI system be able to order the model, get it delivered, unpack it, understand the instructions, and build it?
The 2025 International ATAX Conference, hosted by UNSW Business School
“Back in late 2022, the average prediction was January 2039. But after ChatGPT's release, that shifted dramatically – now the average forecast is May 2030, just five years away. That means an AI system could soon pass an adversarial Turing test and demonstrate complex physical reasoning and fine motor control,” continued Prof. Alarie.
However, what is likely to follow AGI is potentially even more transformative. Superintelligent AI – systems capable of Nobel Prize-level discovery – could emerge within just 30 months of that milestone. “We're talking about something that is at the tip-top of human capabilities in terms of fundamental discovery, innovation, and research,” said Prof. Alarie. “Within the next 10 years, it's likely to be the case that AI is capable of doing virtually any of the things that we are capable of doing, at least as intellectual tasks.”
For the world of tax, this means AI is going to have wide-reaching implications in how tax functions, legal interpretations, and tax liabilities are addressed across different jurisdictions. “I think this has really significant implications for tax professionals and how we work with the technology. It's really why I'm excited to be doing the work that I'm doing,” said Prof. Alarie.
From prediction to real-time tax research
Initially focused on predictive modelling, Blue J has trained early AI models on datasets of court rulings to forecast how cases would be decided. These AI-powered tools can predict outcomes – such as whether a transaction constituted a trade or business – with over 90% accuracy. “When we started the business 10 years ago, I thought, I know what I'll do, I will publish machine learning predictions about how a bunch of pending [court] cases are likely to go. And we didn’t get any wrong.”
Read more: How AI is reshaping the role of tax professionals
By 2023, the rise of ChatGPT and the widespread adoption of natural language interfaces prompted Blue J to pivot toward real-time, conversational tax research. “ChatGPT exploded... It became clear that we could leverage large language models to deliver better tax research capabilities at Blue J.”
The platform now integrates content from trusted providers like Tax Analysts and KPMG, using these data sources to minimise hallucinations and deliver more accurate legal guidance. With usage rising rapidly, Blue J has answered close to 300,000 questions in March 2025 alone. “We're approaching a disagreement rate of 0.1%. We take that feedback and feed it back into the system to fine-tune the algorithms and improve the outputs.”
How tax professionals can benefit from AI
So what should tax professionals do in response? To benefit from the digital transformation, Prof. Alarie stressed that tax leaders need to move beyond talk and start actively implementing AI.
“The lived reality, I suspect, inside a lot of tax administrations – it’s just very hard. Culturally, there's so much of the day-to-day business that is not innovating,” he said. “These technologies are so effective that there's a real missed opportunity if there's a disconnect between the public rhetoric and what's actually happening in the tax administration.”
Subscribe to BusinessThink for the latest research, analysis and insights from UNSW Business School
His advice for tax professionals: “Start with things that have discrete elements. Run them as pilot programs. Be versatile and scale iteratively.”
He also encouraged organisations to invest in data foundations, data processing infrastructure, and internal AI capability. “I think it's a no-brainer to strengthen the data foundations. So do the work to organise your data as a tax administration and then start investing in building the internal AI expertise.”