Automation Is Not the Same as Accountability
That is why I get concerned when I hear that AI can simply take over accounting and handle the month-end close.
If closing the books were only a technology or math problem, that argument would be much easier to accept. AI can automate reconciliations, propose journal entries, identify unusual transactions, and accelerate financial reporting. Those are useful capabilities.
But this isn’t really an accounting problem, is it?! It is also a people problem.
A month-end close connects departments, confirms what actually happened, and puts financial decisions through independent review. It requires someone to ask whether a service was delivered, whether an obligation has been disclosed, or whether a transaction reflects its underlying business purpose. With added non-profit complexity: did it conform to grant and contract requirements? Were things allowable? Is the allocation methodology in the journal entry appropriate?
A balanced journal entry cannot answer all of those questions.
Why Multiple Sets of Eyes Still Matter
Multiple sets of eyes serve a purpose.
Consider a program manager who knows a milestone was missed while accounting has an invoice suggesting the work is complete. Or a department head who has committed to a significant expense that has not yet reached accounts payable.
AI might identify the discrepancy if it has access to the right evidence. It cannot independently establish facts that people have withheld. It also cannot guarantee that someone will tell the truth.
This is why responsibilities are divided. Ideally, non-profit program staff initiates a transaction, another approves it perhaps compares it to the grant agreement, and someone else in admin/finance reconciles or reviews the results. Those controls create opportunities to catch errors, challenge assumptions, and detect fraud or public funds misuse.
Replacing that process with non-human automated workflows can remove the very structure that was protecting the organization.
Having an AI agent check another AI agent also requires scrutiny. If both rely on the same incomplete records, flawed assumptions, or compromised access, agreement between them provides limited assurance.
Now, here is another thought, as AI is becoming the “all in one” solution.
When Oversight Fails: Lessons from Enron and WorldCom
Remember the scandals surrounding Enron and WorldCom? It helped expose the consequences of weak oversight, executive pressure, and misleading financial reporting.
At WorldCom, improper accounting included moving operating expenses into capital accounts, making financial performance appear stronger. Former CEO Bernard Ebbers denied knowing about the fraud and sought to distance himself from accounting decisions; he was nevertheless convicted of conspiracy, securities fraud, and false regulatory filings.
At Enron, the SEC alleged that senior executives manipulated reported results and concealed losses. Its case against Kenneth Lay specifically alleged that he received internal warnings and financial information while continuing to make misleading public statements. The problem was much deeper than whether the arithmetic worked.
It would be inaccurate to say executives had a legal free pass before Sarbanes–Oxley simply by claiming they had not reviewed the financial statements. Fraud was already illegal. But the scandals made the distance between executive authority and claimed financial ignorance impossible to ignore.
The Sarbanes–Oxley Act of 2002 made personal accountability more explicit. Under its Section 302 certification requirements, the principal executive and financial officers of SEC reporting companies must certify that they reviewed the reports and accept specified responsibilities for disclosure controls and internal control over financial reporting.
The principle matters well beyond public companies: leadership must understand the financial information it presents and the process that produced it.
Will “AI Did It” Become the Next Excuse?
The “AI did it” is the next version of the same excuse, I am calling it, just mark my words…
After a similar scandal as the Enron example, my prediction is that future requirements will address AI’s role in financial reporting more explicitly.
I expect greater demands to document where AI was used, what it could access or change, who reviewed its work, and who authorized the final result. That is my prediction about the direction of AI governance in accounting and financial reporting.
AI Can Do the Work. Leadership Must Own the Results.
The accountability question will be familiar: did leadership exercise meaningful oversight, or did it accept a convenient output without asking enough questions?
The “AI did it” should not become a way to avoid answering. To avoid thinking. To avoid being inquisitive. To avoid creativity.
Or, to avoid accountability.
