By Kevin Boeckholt, CPA, Accordance.

Following the IRS Office of Professional Responsibility's release of Alert 2026-19, Introductory Guidelines for Responsible AI Use in Federal Tax Practice on June 24, 2026, much discourse has focused on one critical point: billing practices under section 10.27(a). This section stipulates that charging clients for manual research time that AI has effectively replaced might be deemed an unconscionable fee.

While this is indeed a pressing compliance issue, it overlooks what should be the core focus for firms assessing their AI tools—transparency, not cost. Per section 10.37, tax practitioners must recognize that depending on an AI’s output might be seen as unreasonable if its logic cannot be traced back to a trustworthy source. The opacity of AI systems is now the real metric that firms need to consider when evaluating these technologies. Without clarity, the dialogue around billing remains unresolved.

The alert predominantly reiterates past Circular 230 obligations while adapting them to the evolving landscape of AI. Practitioners still bear responsibility for AI-generated content; erroneous citations can lead to sanctions, and client data must remain secure. The essence remains unchanged, but the application in the context of AI introduces new challenges, particularly around the opacity issue.

Opacity as a Legal Risk

Most mainstream AI systems, particularly the language models emerging in tax and accounting sectors, struggle with reasoning transparency. These AI tools can produce coherent and confident answers but frequently fail to accurately reference the specific legal statute, regulation, or case law underlying their conclusions. A striking instance highlighted in the alert involved a report drafted by Deloitte Australia for the government, which included fabricated citations and made-up quotes attributed to a judge—all stemming from an unverified AI process that went unnoticed until an external academic raised the alarm. Deloitte subsequently issued partial refunds, illustrating a critical failure in oversight and accountability.

This incident underscores a significant concern: the failure of these AI systems remains hidden until someone meticulously checks the sources. A tax memo can easily fall prey to similar issues, displaying the same kind of risk even without extensive scrutiny from external parties. It’s a quiet pitfall, yet firms could have flawed reasoning sitting unexamined in client files, waiting to be discovered.

Two critical provisions amplify the necessity for diligence. The alert reinforces section 10.22, which mandates that practitioners verify the accuracy of all facts and citations pre-delivery, and section 10.35, which insists that practitioners grasp how an AI system composes its outputs and judge whether the results meet IRS standards. Truly understanding AI output necessitates more than glancing at a vendor’s promotional materials; it requires in-depth knowledge about the origin of each factual assertion and its legality. Thus, professionals must independently validate any claims from the AI, essentially reverting to the research methods they sought to replace.

When AI systems fail to provide clear citations, firms will need to reconstruct those links manually. This effectively doubles the workload that the AI was intended to streamline, leading preparers to incur costs for both producing and afterwards validating the research claimed by the AI. They find themselves in a situation where they’re effectively paying twice for the same due diligence.

Defining Traceability in AI

It is vital to clarify what differentiates compliant AI tools from those that aren’t. A conventional AI system, when queried about tax legislation, may yield a polished paragraph with an assertive conclusion. If it references any legal sources, they appear simply as text items like case names or regulation citations. Yet, this output fails to allow preparers to confirm either the existence of those citations or their content’s accuracy without conducting independent research. Trust in such citations relies solely on the AI’s assurance, precisely the risk the alert cautions against.

In contrast, a traceable system maintains each factual claim’s integrity by linking directly to a corresponding source. Each piece of output should offer access to specific statutes, rulings, or regulations in their original form, enabling professionals to verify claims without taking the AI’s interpretation at face value. This fundamental change alters how verification duties under section 10.22 should be approached; instead of starting from scratch, the task becomes to confirm the presented source’s reliability, significantly reducing the burden that is otherwise expected from a professional.

This is more than a procedural distinction—it's an engineering challenge that can be addressed. Some newer systems effectively segregate the components that gather primary research from those that synthesize responses, ensuring that information presented is strictly derived from validated research records. Consequently, each output can include tagged markers denoting their specific sources. At the point of drafting, any discrepancies between the AI output and the sourced material can be systematically eliminated. This distinction transforms how firms can engage with their AI tools and meet the scrutiny established in Circular 230.

Why Billing Cannot Be the Initial Focus

This brings us back to the pivotal discussion around section 10.27(a), which has become the focal point of many discussions. The guidance suggests that firms should transparently pass on any cost savings gained from AI efficiencies and ideally move toward value-based billing. While this direction seems sensible in the long run, prioritizing it at this moment risks overlooking critical preceding issues.

A firm can't accurately establish a value-based price or substantiate savings owed to clients if the tool generating the work lacks traceability. You cannot charge clients for results you cannot verify, and without traceability, reliability comes into question. Opacity must be addressed before any billing practices can be refined or revised.

Action Steps for Firms

Alert 2026-19 centralizes a fundamental question in AI tool evaluation: Can the system substantiate its answer with a verifiable primary source, or does it merely provide polished output? Firms do not need to wait for further directives from the OPR to deliberate and act upon this vital inquiry.

During any evaluation or renewal of AI applications, firms should pose direct queries to vendors about output provenance. Specifically, can the system trace each substantial claim back to a primary source accessible for inspection? Moreover, documenting which claims have been validated against those primary sources and by whom will become indispensable should the work come under scrutiny. This documentation aligns perfectly with due diligence and competence mandates outlined in sections 10.22 and 10.35.

Finally, firms should recognize the verification process as a genuine aspect of using opaque tools. They should candidly account for this time rather than presuming that any advertised efficiency gains are fully realized after human review is incorporated.

Ultimately, much of what the alert discusses—verification workload, billing integrity, and exposure under section 10.37—stems from a singular question regarding traceability. An AI tool engineered for tax research must inherently document its processes against primary sources to satisfy regulatory standards. Without this level of transparency, firms face untenable risks in an increasingly complex environment.

===

Bio: Kevin Boeckholt, CPA, leads Accordance‘s innovation practice, collaborating with tax and accounting professionals to implement AI effectively. With prior experience in product development at an early-stage tax AI firm and in Tax Technology Consulting at Deloitte, he understands the intricate realities facing the industry.

Looking Ahead: Navigating Compliance in AI Advisory

As we move into an era defined by artificial intelligence, the implications for tax advisory firms are both vast and complex. While AI presents extraordinary opportunities for efficiency and insight, it also raises significant compliance concerns, particularly around adherence to Circular 230 regulations. The challenge lies not just in integrating advanced tools but ensuring that they align with the regulatory frameworks demanding integrity and accuracy in tax advisory practices. What poses the most risk, however, seems to be the potential for oversight regarding billing practices. Firms adopting AI technologies need to question if their usage complies fully with existing standards. It’s not just about the technical capabilities; it’s about how well firms can adapt their operational protocols to mitigate risks. AI’s role in automating processes is enticing, but the possibility of non-compliance due to lack of oversight is a real pitfall. If you’re working in this space, prioritize a thorough understanding of AI's limitations within your business model. Make sure your systems are capable of not just producing results but also documenting them in a way that aligns with regulatory scrutiny. You may very well need to innovate your compliance protocols alongside your technology use. The current conversation doesn't address all the uncertainties lingering over AI deployment, especially its evolving risk profile. As the IRS and other regulatory bodies refine their positions on these technologies, tax professionals must remain vigilant. Balancing innovation while maintaining strict compliance isn't just prudent; it’s necessary for the sustainable success of any advisory firm. Ultimately, being proactive now about these concerns may safeguard your practice from future complications. Keeping an eye on regulatory updates and fostering a culture of compliance within your firm will be essential as the landscape shifts. This isn’t just a passing trend; it’s a vital part of the evolution of tax advisory.