Artificial intelligence is becoming more than a tool for drafting emails, summarizing documents, and answering questions. Emerging AI systems can coordinate tasks, interact with applications, and initiate actions within established workflows. For tax and accounting professionals, this shift raises a critical question: Who is accountable when AI moves from recommending an action to performing it?
The distinction matters across tax preparation, bookkeeping, financial reporting, advisory services, and internal audit. An AI tool that summarizes a client’s records creates different risks from one authorized to change classifications, update information, or send communications.
The central challenge is not simply adopting AI. It is ensuring that the authority granted to technology is matched by appropriate oversight.
AI assistance is not the same as AI authority
Consider three possible arrangements.
In an assisted workflow, AI prepares information or recommends a course of action. A professional evaluates the recommendation and decides what happens next.
In a bounded workflow, AI performs predefined activities within approved limits, such as categorizing routine items or routing exceptions for review.
In a more autonomous workflow, AI may coordinate multiple steps and execute authorized tasks across connected systems without requiring approval at every stage.
These arrangements require different safeguards. As operational authority increases, professionals should ask four questions before relying on the system.
1. What is the system authorized to do?
A tax application might identify a possible inconsistency in client information. That does not mean it should independently change a return or transmit information to a taxing authority.
Likewise, an accounting application might propose a journal entry without being authorized to post it.
Organizations should document which actions AI may recommend, which it may execute, and which require professional approval. Higher-consequence activities deserve more restrictive permissions and stronger review procedures.
The governing question is not what the technology can do, but what it should be permitted to do.
2. Can we reconstruct what happened?
When an AI-enabled process produces an unexpected result, professionals need more than a record showing that the software was used.
They should be able to determine what action occurred, when it occurred, what information was available, and which permissions or procedures governed the activity.
For example, if an automated process changes a transaction classification, the accounting team should be able to identify the change, investigate its basis, and determine whether correction is necessary.
Activity logs, approval histories, exception records, and documented workflow changes support this review.
Complete technical explanations of every AI output may not always be available. However, sufficient operational documentation should exist to investigate consequential actions and establish an accountable record.
3. Can a professional intervene?
Human oversight is effective only when someone has the authority, information, and opportunity to act.
A manager who receives a notification after an irreversible action has occurred may have visibility but little meaningful control.
Tax and accounting practices should determine when human approval is required, which exceptions trigger escalation, and how an automated process can be paused or corrected.
A routine administrative task may be suitable for periodic review. A proposed tax filing, material accounting adjustment, or consequential client communication may warrant approval before execution.
The objective is not to eliminate automation. It is to preserve professional judgment where the consequences require it.
4. Who is responsible for the outcome?
Delegating a task to AI does not eliminate organizational responsibility.
Before implementing an AI-enabled workflow, firms and finance departments should identify who authorizes its use, who monitors its performance, who investigates errors, and who can suspend the process.
Responsibility may be distributed among engagement leaders, tax managers, controllers, internal auditors, and technology personnel. But it should never be left undefined.
Clear ownership also helps professionals respond consistently when clients question an outcome or when an internal review identifies a problem.
Start with one workflow
A practical first step is to select one existing AI-enabled process and document four elements: its permitted actions, available records, intervention points, and responsible personnel.
Then consider a realistic failure scenario. If the system made an incorrect classification, initiated an unauthorized communication, or overlooked an exception, could the organization reconstruct the event, limit further consequences, correct the result, and identify who should respond?
If those questions cannot be answered, expanding the system’s autonomy may be premature.
For tax and accounting professionals, effective AI governance does not require resisting technological progress. It requires recognizing that assistance, delegation, and autonomous execution create different accountability demands.
As AI receives greater authority to act, the safeguards surrounding its decisions and actions must become correspondingly stronger.
ABOUT THE AUTHOR:
Juan Ojeda Jr., MSA, MBA, is a tax professional, strategic advisor, independent researcher, and founder of Structural Accountability. His background spans tax compliance and controversy, public accounting, practice operations, internal controls, and organizational governance. His article, “When the Client Is the Pressure: Navigating Circular 230 Under Real-World Demands,” appeared in the Spring 2026 edition (Issue 44) of the National Association of Enrolled Agents’ EA Journal. His independent research and preprints are available through SSRN. He also publishes the Structural Accountability monthly LinkedIn newsletter and audio briefs distributed across podcast platforms, examining tax administration, financial reporting, AI oversight, and professional accountability.
Sign in to get access to this free resource, and all of our whitepapers and reports.
Download this content today!
Register Now Already registered? Click here to Log In
