Please click here to view Danny Werfel’s AI Risk Framework for Tax Authorities and Tax Preparers.
A commercial airline pilot recently told me that at some point in my lifetime passengers will fly on aircraft with no pilot in the cockpit.
"Not this passenger," I responded, pointing at myself with my two thumbs.
It wasn't because I doubted the technology. I can be convinced, as a technical matter, that autonomous aviation will one day be safe enough to completely eliminate the need for a human pilot. I assured the pilot he didn't need to prove it to me mathematically. I believe the math. My hesitation comes from a different place.
Trust is more than a calculation. It is influenced by qualitative considerations that resist easy measurement, including subjective expectations we may have on norms, traditions, and the as yet unresolved tension between person and machine when it comes to judgement.
My conversation with the pilot captures a divide that I increasingly see in discussions between the sellers and buyers of AI.
Those selling AI understandably focus on what the technology can do. Those responsible for deploying it, particularly in highly regulated professions, focus on a different question: “What gives us permission to trust it?”
Few professions feel that tension more acutely than tax, where AI solutions are multiplying.
Responsible tax professionals are rarely rewarded for being first. They are rewarded for being right. So, no matter how enticing the promised productivity seen in an AI demonstration, a persistent question will stall adoption until it is answered: “How do I know this system is reliable enough to use?”
In mature markets, the buyer doesn’t start with a blank sheet of paper. Take, for example, a government defence authority considering the purchase of an airplane from Boeing or Lockheed Martin. There are many questions that the buyer need not ask. The buyer will know, for example, that the fuel system, wiring and rudder were all built to recognised engineering standards and heavily tested and inspected before it leaves the assembly line and moves into the marketplace.
AI offers no comparable foundation.
Organisations purchasing AI today cannot assume that common safeguards exist. There is no universally accepted engineering standard, no independent inspection regime, and no agreed-upon baseline that tells a buyer what questions have already been answered – or which ones still need to be asked.
Ironically, the absence of regulation in this scenario is a barrier to innovation.
Regulation is often portrayed as something that slows risk-taking, innovation and technological progress. In the case of AI, the opposite can also be true. Without mature standards, every tax authority, accounting firm, and corporate tax department must conduct a fair amount of the diligence that regulation would ordinarily perform on society's behalf. For a profession built on managing risk, that uncertainty naturally slows adoption.
That is precisely why I developed the accompanying AI Risk Framework for Tax Authorities and Tax Preparers. The framework is not intended to replace future regulation. Nor is it an attempt to prescribe a single approach to AI governance. Instead, it is designed to bridge today's gap between rapid technological advancement and the slower pace of regulatory development.
It provides tax organisations with a structured way to identify, prioritise, and mitigate the risks unique to AI in tax, and create a practical permission structure for innovation while governments, professional bodies, and standards organisations continue developing broader guardrails.
The question facing the tax profession is no longer whether AI can transform our work. It already is. The more pressing question is whether we have developed enough confidence in the governance surrounding that technology to fully embrace its promise.
Until regulation and standards catch up, that responsibility rests with each of us. My hope is that this framework helps organisations ask the right questions – not to slow innovation, but to make responsible innovation easier.
