In brief
ACCA surveyed 1,178 finance and accounting professionals across 135 countries.
AI use was selected as a leading ethical challenge by 61%, up from 32% in 2024.
72% still said AI improves the quality and efficiency of their work.
AI is useful. That does not make it easy to govern.
Accountants are not choosing between loving AI and distrusting it. Many are doing both. In ACCA's Navigating Ethics for Trust research, 61% of respondents identified AI use as a leading ethical challenge, while 72% said the tools improve work quality and efficiency.
The research draws on 1,178 finance and accounting professionals across 135 countries, plus roundtable discussions. It describes the views of those respondents, not a measured rate of misconduct across every accounting firm.
The concern has moved fast
AI was selected by 32% in the 2024 research. By 2026, it had moved ahead of ethical leadership and ethical culture, which were selected by 43% and 39%, respectively.
Those categories overlap. A tool can produce a weak answer, but an organization decides whether anyone checks it, who can challenge it and what happens when a deadline is tight. The technology risk becomes a management risk quickly.
ACCA highlights familiar pressure points: overreliance on outputs, weak verification, unclear responsibility, confidentiality risks and difficulty explaining how a conclusion was reached. A convincing answer is not necessarily a sound one.
Keep a human name on the conclusion
Roundtable participants stressed that responsibility stays with people. AI-assisted work still needs review, challenge and documentation. Staff should be able to explain the conclusion without pointing to a chatbot as the authority.
That suggests a simple test for a firm policy. Who owns this output? What source evidence supports it? What client information entered the tool? Who reviewed the result before it left the firm? If those questions have no answer, buying another license will not solve the problem.
The response is not to ban every useful tool. It is to define approved uses, protect confidential information and require checks that fit the risk of the assignment.
Training must go beyond better prompts
In the survey, 89% said they apply professional skepticism and ethical judgment when using AI. That is encouraging, but it is self-reported behavior, not proof that every output was checked.
A practical training session should use flawed answers, missing context and unsupported conclusions, not only polished demonstrations. Let staff practice finding the error and documenting the correction.
The opportunity is real: faster work and better support for analysis. The responsibility is just as real. AI can help produce the work; the firm still has to stand behind it.
