Artificial intelligence can make a reporting team look more efficient very quickly.

A rough paragraph can become polished in seconds. A lengthy document can be summarised. Two versions of an annual report can be compared. Disclosures can be checked for inconsistencies and technical material can be turned into clear-looking commentary.

Those are genuine advantages.

They also create a problem that is easy to underestimate.

Bad judgement does not always look bad when AI has rewritten it.

A weak assumption can be expressed confidently. An incomplete analysis can sound comprehensive. A questionable accounting conclusion can be presented in language that appears professional enough to discourage further challenge.

That may prove to be one of the most important corporate reporting risks created by generative AI.

The problem is not simply that AI can make mistakes. People make mistakes too.

The problem is that AI can make a mistake look finished.

For businesses, that raises questions about review, governance and professional accountability. For ACCA SBR candidates, it creates a strong current reporting issue because it connects technology with judgement, ethics, internal control and the quality of financial information.

Candidates developing their technique with an ACCA SBR tutor should be able to move beyond the basic argument that AI is faster but less reliable. The more interesting question is what happens when speed and polished presentation make an unreliable conclusion more difficult to recognise.

Corporate reporting is an obvious place to use AI

Preparing an annual report involves a significant amount of repetitive work.

Reporting teams compare current disclosures with the previous year. They check figures against supporting schedules. They update standard wording, reconcile terminology and review documents for inconsistencies.

They also spend time turning information from different parts of the organisation into readable narrative.

AI can help with much of this.

Recent UK research into corporate reporting shows that generative AI is already being used for activities such as drafting narrative sections, copyediting, identifying reporting trends, checking consistency and reviewing tone.

That makes sense.

These are tasks where technology can reduce manual effort without necessarily making the final reporting decision.

A first draft of a straightforward disclosure can be produced more quickly.

A lengthy accounting paper can be summarised for review.

A system can identify that the strategic report uses one performance measure while the financial statements use another.

It can compare this year’s report with last year’s and flag wording that has not been updated.

None of that requires the machine to take responsibility for the accounting judgement.

The danger begins when efficiency starts to create trust.

Good writing can disguise weak thinking

Imagine two accounting papers.

The first has been written manually by a junior accountant. The analysis is uncertain. The grammar is poor. Several sentences are difficult to follow and the conclusion appears weak.

A reviewer immediately knows the paper needs attention.

The second paper contains exactly the same underlying reasoning, but it has been rewritten by generative AI.

The sentences are clear.

The structure looks professional.

The conclusion sounds confident.

Which one receives more challenge?

Potentially, the first.

That is the uncomfortable part.

Presentation quality and analytical quality are not the same thing, but humans naturally use presentation as a signal.

A confident-looking document can create an impression of competence before the reviewer has tested the reasoning.

AI strengthens that effect because it is very good at producing language that sounds complete.

This means finance teams need to become more disciplined about separating the quality of the writing from the quality of the judgement.

A polished answer still needs evidence.

The highest-risk areas are the ones involving judgement

Using AI to correct spelling in a draft disclosure creates relatively little reporting risk.

Using it to support an impairment conclusion is very different.

High-judgement areas can include:

  • impairment forecasts
  • going concern assessments
  • provisions and contingencies
  • expected credit losses
  • fair value estimates
  • useful economic lives
  • revenue recognition judgements
  • climate-related assumptions
  • narrative explanations of performance

These areas contain uncertainty.

There may be several reasonable assumptions and conflicting pieces of evidence.

Management must decide which information deserves the most weight and whether the final conclusion faithfully represents the company’s circumstances.

AI can assist with the analysis, but it cannot remove that uncertainty.

If anything, a neat output may make uncertainty less visible.

That creates a governance problem.

Consider an impairment review

Suppose a company is testing a cash-generating unit for impairment.

Recent sales have fallen by 12 per cent.

Management expects a strong recovery over the next three years and has prepared a forecast showing revenue growth well above recent experience.

The finance team asks an AI tool to review the forecast assumptions and draft an impairment paper.

The resulting document explains that management expects revenue to recover as market conditions improve. It refers to strategic initiatives, improving efficiency and the company’s long-term growth prospects.

It reads well.

But what evidence supports the recovery?

Has the company gained new customers?

Have market forecasts improved?

Has recent trading shown the beginning of a turnaround?

Were previous management forecasts accurate?

If those questions are not answered, the polished narrative is irrelevant.

The AI has not necessarily created the bad judgement. Management may already have supplied optimistic assumptions.

What it has done is package those assumptions in a way that may make them feel more credible.

That is why human review cannot focus only on whether the report makes sense.

The reviewer must challenge what sits underneath it.

AI may increase confirmation bias

Management bias existed long before artificial intelligence.

A board may prefer a forecast that avoids an impairment.

A finance director may favour a revenue treatment that improves the year’s result.

Management may want an annual report to present a difficult trading year as a temporary setback.

AI can unintentionally strengthen those preferences.

The result depends partly on the question it is asked.

If a user asks:

“Explain why management’s revenue forecast remains reasonable despite the recent decline.”

the system is already being directed towards supporting the forecast.

A very different analysis might result from asking:

“Identify evidence that could indicate management’s forecast is overly optimistic.”

This matters because users may treat AI as if it were an independent opinion.

It is not.

The prompt can shape the answer.

The information supplied can shape the answer.

Missing information can shape the answer.

If management provides only evidence that supports its preferred conclusion, the resulting output may simply become a more polished version of management bias.

The reviewer needs to challenge the inputs

Traditional review often focuses on the final document.

With AI-assisted reporting, reviewers may need to move further upstream.

They should understand what information the system received.

If an AI-generated paper recommends that no impairment is required, the reviewer should ask which forecasts, budgets, market data and assumptions were provided.

Was contradictory evidence included?

Was recent underperformance included?

Was an unsuccessful previous forecast included?

Was the system told that management bonuses depend on maintaining a particular profit level?

Without those facts, the apparent quality of the analysis means very little.

AI cannot evaluate evidence it never receives.

That creates an important internal control point.

The organisation should not simply review the output. It should understand the process used to create it.

A fast answer can shorten the thinking process

There is another risk that has less to do with technical accuracy and more to do with behaviour.

When producing a difficult accounting paper manually, the writer has to think through the issue.

They read the contract.

They find the relevant accounting requirement.

They compare different interpretations.

They draft the conclusion.

That process can reveal uncertainty.

AI can move directly from information to apparent answer.

That saves time, but some of the thinking may disappear with it.

A junior accountant who receives a convincing draft within seconds may be less likely to spend an hour questioning whether the conclusion is correct.

This is sometimes described as automation bias.

People begin to assume that the system is probably right.

In corporate reporting, that is dangerous because many important decisions do not have a simple factual answer.

Professional judgement cannot be automated merely by making the drafting process faster.

Human oversight only works when the human actually challenges

Many organisations respond to concerns about AI by saying there will always be a human in the loop.

That sounds reassuring.

It is only useful if the human review is meaningful.

A reviewer who scans an AI-generated paragraph and clicks approve is technically a human in the loop.

That does not make the control effective.

Human oversight should involve challenge.

The reviewer should understand the issue well enough to identify a weak conclusion.

They should know which source information supports it.

They should look for contradictory evidence.

They should be willing to reject the output entirely.

Recent corporate reporting research suggests that companies recognise the importance of human oversight, but formal requirements are not yet universal.

That gap matters.

An organisation may believe its AI use is controlled because somebody reviews the result. In reality, the review may be little more than proofreading.

Different uses need different controls

It would be excessive to require board-level approval every time somebody uses AI to improve a sentence.

Controls should match the risk.

Using AI to correct grammar in a routine disclosure is a lower-risk activity.

Using it to draft an accounting policy may require greater technical review.

Using it to analyse a significant judgement such as going concern or impairment should involve stronger control again.

A proportionate approach might consider:

the materiality of the issue, the level of judgement involved, the reliability of the source data, the confidentiality of the information and the consequences of an incorrect conclusion.

The greater the risk, the less suitable it becomes to rely on an AI output without detailed professional review.

This allows businesses to benefit from efficiency without pretending every use of AI is equally safe.

Data quality becomes even more important

An AI system cannot repair poor source information simply because the output is sophisticated.

If the underlying data is incomplete, the answer may be incomplete.

If the data is wrong, the analysis may be wrong.

If different departments use inconsistent definitions, AI may combine information that should not have been combined.

This is particularly important in areas where businesses are collecting newer forms of information, such as sustainability data.

Financial reporting systems often have established controls built over many years.

Non-financial data may come from spreadsheets, operational systems, suppliers and manual submissions from different locations.

Using AI to process that information more quickly does not solve weaknesses in how it was collected.

It may make those weaknesses harder to see because the final report appears organised.

Good reporting still starts with reliable data.

The annual report can become consistently wrong

One advantage of AI is its ability to create consistency.

It can align terminology across a long document.

It can rewrite sections in a common style.

It can identify contradictions between different disclosures.

Those are useful capabilities.

There is also another possibility.

AI can make an incorrect message consistently wrong across the whole report.

Imagine management believes that poor performance is entirely caused by temporary market conditions.

The same assumption is then used when AI drafts the strategic report, risk commentary and outlook statement.

The annual report may become extremely consistent.

It may still be misleading if internal execution problems were an equally important cause of the decline.

Consistency is therefore not enough.

The story needs to be supported by evidence.

A connected report should tell one accurate story, not merely one repeated story.

Authenticity is becoming more important

One of the more interesting findings from recent UK reporting research is the importance investors continue to place on authenticity.

That makes sense.

Investors want to understand what management actually thinks.

They want to know why performance changed, what risks concern the board and which decisions management is making.

Generic corporate language already weakens reporting.

AI could increase that problem if companies begin producing similarly polished explanations.

The annual report should still sound like it belongs to the organisation.

If every difficult year is described as a period of “resilience in a challenging environment”, the language tells users very little.

Management needs to explain the actual challenge.

Did customers buy less?

Did costs rise faster than prices?

Did a product launch fail?

Did integration problems damage margins?

Did a regulatory change delay investment?

AI can help organise the wording.

It should not sanitise the underlying reality.

Confidentiality adds another risk

Corporate reporting teams often work with information that has not yet been made public.

Draft results, impairment forecasts, restructuring proposals, acquisition plans, legal disputes and board papers may all form part of the reporting process.

Using public AI tools without appropriate controls can create confidentiality and data security concerns.

Recent research indicates that enterprise AI tools dominate corporate use, but some users are also using public tools.

That should concern boards.

The organisation may have a carefully controlled reporting process while an employee copies sensitive information into an external system because it makes drafting faster.

A proper policy needs to address which tools can be used and what information can be entered.

Professional accountants also have ethical responsibilities regarding confidentiality.

Convenience does not remove those obligations.

Boards may know less about AI use than they think

Another interesting risk is the gap between formal adoption and informal use.

Senior management may say that AI is not yet used significantly in corporate reporting.

Junior staff may already be using it for first drafts, summaries, checking or technical research.

Both statements can be true.

AI tools are easy to access and often embedded in software employees already use.

That means boards cannot rely only on formal implementation projects to understand their exposure.

They should ask how staff are actually using AI.

Which tools?

For which tasks?

With what data?

Under whose supervision?

A policy written before real usage was understood may miss the actual risk.

Governance has to follow behaviour.

Auditors face the same problem

The challenge does not end with preparers.

Auditors are also using technology to process information, identify anomalies and support documentation.

The same risk applies.

An automated analysis may look objective because it was produced by a system.

The audit team still needs to understand it.

If AI identifies a group of transactions as low risk, the auditor must know why.

If it summarises a contract, someone must make sure the summary did not omit a clause that changes the accounting.

If it assists with reviewing management’s forecasts, professional scepticism remains essential.

Technology may increase the amount of information an auditor can examine.

It does not transfer responsibility for the audit opinion.

Why this matters for SBR

This is a useful SBR issue because it crosses several parts of professional reporting.

A scenario could involve AI being used to prepare a disclosure.

It could involve an impairment model.

It could involve confidential information being entered into an unauthorised system.

It could involve management accepting AI-generated accounting advice without sufficient challenge.

It could involve an audit committee that has no idea how widely the finance team is using the technology.

The weak answer would say:

“AI can increase efficiency but there is a risk of errors.”

That is true, but it is superficial.

A stronger answer identifies the specific failure.

For example:

Management has relied on AI-generated impairment analysis based solely on internally prepared forecasts. The conclusion may be biased because recent underperformance and external market evidence have not been considered. The finance director should require an experienced reviewer to challenge the assumptions against actual results and independent evidence before the impairment conclusion is approved.

That sounds like professional advice.

It identifies the risk, applies the facts and recommends an action.

Professional scepticism becomes more valuable

The growth of AI does not reduce the value of accountants who understand financial reporting.

It increases the value of people who can challenge an answer.

When information can be produced quickly, judgement becomes the scarce skill.

Can you tell whether the answer is credible?

Can you identify the assumption driving the conclusion?

Can you spot the missing evidence?

Can you recognise when technically accurate language produces a misleading overall impression?

Can you explain what should change?

Those are professional skills.

They are also exactly the kind of skills candidates should develop through SBR question practice.

How candidates should practise this topic

Do not try to memorise a generic AI essay.

Instead, practise applying a few principles to different scenarios.

Start with accountability.

Who owns the decision?

Then consider reliability.

What information supports the output?

Then consider judgement.

Which assumptions need professional challenge?

Then consider governance.

Who reviews and approves the result?

Finally, consider ethics.

Has confidential information been protected and has the accountant exercised appropriate competence and care?

The answer should always follow the facts.

If you want a structured way to develop this type of applied writing alongside the wider syllabus, an ACCA SBR course can provide regular question practice and feedback. The important point is that candidates practise producing professional responses rather than simply learning lists of AI risks.

The opportunity should not be lost

None of this means businesses should avoid AI.

The efficiency opportunity is substantial.

Reporting teams spend huge amounts of time on repetitive work that technology can perform more quickly.

AI can help identify inconsistencies.

It can reduce drafting time.

It can support benchmarking.

It can help users navigate complex technical material.

It can give finance professionals more time to investigate the issues that genuinely require judgement.

That is the ideal outcome.

Technology handles more of the routine work while people spend more time thinking.

The danger is allowing the opposite to happen.

If AI handles the drafting and humans reduce the thinking because the output already looks convincing, reporting quality could deteriorate while appearing to improve.

That is why the quality of human judgement remains central.

What boards should take from this

The important question for a board is not whether the organisation uses artificial intelligence.

It probably already does somewhere.

The better questions are where it is being used, what decisions it influences and whether the level of review matches the level of risk.

Using AI to produce a first draft can be sensible.

Using it to challenge a long document for inconsistencies can be valuable.

Using it to support analysis can improve efficiency.

The line should be drawn where the organisation stops understanding or challenging the conclusion.

Corporate reporting needs accountability.

Somebody has to own the judgement.

What to do next

AI is likely to become increasingly normal in finance and reporting.

The organisations that use it well will not necessarily be the ones that automate the most.

They will be the ones that understand which activities are suitable for automation and which require experienced human challenge.

For SBR candidates, the same principle applies.

Do not treat AI as a technology essay.

Treat it as a professional judgement problem.

Ask what evidence supports the answer.

Challenge the assumptions.

Consider the quality of the data.

Identify who remains accountable.

Then recommend a practical control.

AI can make corporate reporting faster.

The profession’s job is to make sure it does not also make bad judgement harder to see.