When Accountants Use AI: What Is Ethical and What Crosses the Line?
This article will count 0.25 units (15 minutes) of unverifiable CPD. Remember to log these units under your membership profile.
Artificial intelligence is already part of everyday accounting.
Accountants can use AI to draft emails, summarise documents, analyse financial information and help prepare reports. Audit software can scan thousands of transactions and identify unusual items. Automated systems can perform reconciliations and process information much faster than a person could.
These tools can save accountants hours of work. But they also create some very practical ethical problems.
Can you upload a client's financial statements to an AI tool to ask it to write a summary? Can an auditor rely on software that says a transaction is low risk? Should employees know if software is monitoring their transactions? What happens if AI gives the wrong answer and an accountant uses it?
These are not questions for the future. Accountants are dealing with them today.
The issue is not whether accountants should use technology. The real issue is knowing where to draw the line.
Having Access Does Not Mean You Can Use It
Accountants often have access to highly confidential information. This may include salaries, bank statements, tax information, financial statements, customer details and employee records.
Technology makes it extremely easy to copy, upload and analyse this information.
Imagine an accountant receives a client's financial statements and needs to prepare a short report. Instead of reading everything manually, the accountant uploads the statements to a public AI chatbot and asks it to identify the main risks.
Within seconds, there is an answer.
But was the accountant allowed to upload the information? Where will it be stored? How will it be used? Has confidential client information now been shared with an outside system?
The fact that an accountant has access to information does not automatically mean there is permission to use it in any way they choose.
This is also important when personal information is involved. South African accountants need to consider the requirements of the Protection of Personal Information Act, commonly known as POPIA.
A useful rule is simple: access to information is not permission to use it however you want.
AI Can Be Wrong Even When It Sounds Right
One of the dangers of AI is that a wrong answer can sound very convincing.
The wording may be professional. The explanation may seem logical. The answer may look like it was prepared by an expert.
That does not make it correct.
Suppose an accountant asks AI to explain the accounting or tax treatment of a transaction. The tool produces a detailed answer and the accountant includes it in a report without checking it.
Later, the answer turns out to be wrong.
Who is responsible?
The AI tool did not sign the report. The accountant did.
The same applies to auditing. An audit program may analyse thousands of transactions and identify 30 unusual items. This is useful, but the auditor still needs to understand why those transactions were selected and consider what the system may have missed.
Technology can support professional judgement. It cannot replace it.
"The System Said So" Is Not Enough
Automation allows accountants and auditors to perform work that previously took many hours.
For example, software may analyse an entire population of transactions instead of an auditor manually selecting a sample.
This can improve audit quality, but only if the results are used properly.
Imagine an automated system reviews company expenses and gives certain transactions a low-risk rating. The auditor accepts that result and performs no further work. Later, fraud is discovered in some of those transactions.
Saying "the system classified them as low risk" does not remove the auditor's responsibility.
Accountants do not need to become computer programmers, but they do need to understand the tools they rely on. They should know what the system does, what information it considers and where it can go wrong.
How Much Monitoring Is Too Much?
Technology can also be used to monitor employees.
A business may analyse expense claims, transactions and spending patterns to identify possible fraud. There may be a valid reason for doing this.
But where should the monitoring stop?
Technology can make it possible to monitor transactions, communications and even employee activity continuously. The fact that a business can collect all this information does not necessarily mean it should.
The information collected should be appropriate for the purpose.
Preventing fraud may justify certain monitoring. It does not automatically justify collecting every piece of information technology makes available.
Computers Can Also Be Biased
People sometimes assume a computer-generated decision must be fair because computers do not have personal opinions.
Unfortunately, it is not that simple.
Many systems learn from historical information. If that information contains unfair patterns, the system can learn those patterns too.
A fraud detection system, for example, may begin identifying certain types of transactions or employees as higher risk because of patterns in old data. That does not necessarily mean those people are more likely to commit fraud.
Accountants should therefore question why a system is producing a particular result.
An automated answer should never become an excuse to stop asking questions.
Confidentiality Still Applies to AI
Confidentiality is not a new requirement in accounting. What has changed is how easily confidential information can leave the accountant's control.
Most accountants would immediately recognise that sending confidential client information to an unknown person would be inappropriate.
Uploading the same information to an online AI system may not feel as serious, but the confidentiality issue remains.
Before entering information into an AI tool, accountants should know whether the tool is approved, how the information will be handled and whether confidential information is allowed to be entered.
Removing the client's name may also not always solve the problem. Financial information may contain other details that identify a business or person.
The Basic Ethical Rules Have Not Changed
Technology may be changing quickly, but the fundamental ethical principles of accounting still apply.
Integrity means being honest about how technology was used and not presenting an AI answer as more reliable than it is.
Objectivity means not allowing technology providers or automated results to unfairly influence professional judgement.
Professional competence and due care mean understanding the tools being used and checking their work properly.
Confidentiality means protecting client and business information, including when using AI.
Professional behaviour means acting in a way that protects the reputation of the accounting profession.
The technology has changed. These responsibilities have not.
Before You Use AI, Ask a Few Questions
Before using AI or another automated system for accounting work, accountants should ask:
Why am I using this tool?
What information am I giving it?
Am I allowed to share that information?
Do I understand how the tool reached its answer?
Have I checked the result?
What could happen if the answer is wrong?
Would I be comfortable telling the client exactly how I used the technology?
That final question can be especially useful. If you would not be comfortable explaining your actions to the client, there may be an ethical issue that needs attention.
Accounting Firms Need Clear Rules
The responsibility cannot rest only on individual accountants.
Firms should have clear policies explaining which AI tools may be used and what information may be entered into them. Employees should also know when AI-generated work must be checked and reviewed.
Training should not only teach employees how to use AI. It should also teach them when not to use it.
Junior employees should have someone they can approach when they are uncertain. They should not have to make difficult decisions about client confidentiality or technology risks on their own.
The Accountant Still Makes the Final Call
AI can make accounting faster. Automation can reduce repetitive work. Data analytics can help auditors identify problems that may previously have gone unnoticed.
These are real benefits.
But faster does not automatically mean better, and possible does not automatically mean appropriate.
AI does not have a professional reputation to protect. It does not have a relationship with the client. It does not sign the audit report or explain to a regulator why confidential information was shared.
The accountant does.
That is why professional judgement remains so important.
Before uploading the document, accepting the automated result or relying on an AI-generated answer, accountants should ask one simple question:
I can do this, but should I?
Knowing when the answer is yes and when it is no is becoming an important part of being a professional accountant.
