Artificial intelligence (AI) is moving into the daily flow of work faster than many leaders expected. The technology can accelerate research, summarize information, draft communications, and improve efficiency. It can also make it easier to move too quickly, trust too easily, and hand off more judgment than you should.
Mustafa Suleyman, CEO of Microsoft AI, captured that tension in a Financial Times interview: “So white-collar work, where you’re sitting down at a computer [being an accountant], most of those tasks will be fully automated by an AI within the next 12 to 18 months.”
Whether that sounds exciting, alarming, or a little of both, it points to the same reality: AI isn’t just changing tools—it’s changing the context in which professionals make decisions.
The accounting world has seen major technology shifts before. The profession moved from paper ledgers to spreadsheets, from desktop software to cloud platforms, and from static reports to real-time dashboards. Each wave improved capability, but none removed the need for judgment. If anything, each technological shift made human discernment more important.
AI also requires that same human judgment, but with one important twist: It can influence behaviors.
Much of the public conversation around AI focuses on hallucinations, misinformation, bias, data security, and deepfakes. Those concerns are real. Leaders should absolutely be mindful of trusting AI output too readily, misrepresenting AI-powered capabilities, disclosing confidential data, overlooking unfair impact, and sacrificing due care for expediency. But the deeper concern may be even more significant: AI can create conditions in which people begin to abdicate judgment, professional skepticism, and accountability.
That concern becomes more concrete when you look at emerging research on behavior. In a study published by Nature, participants rolled dice and reported the results, with higher numbers producing greater rewards. When people reported directly, only 5% cheated. But when AI entered the reporting process, cheating jumped between 25% and 88% depending on how AI was used.
In a second test involving income reporting and a flat tax, participants were more likely to cheat when reporting was delegated to AI, and AI itself was more likely to cheat when it wasn’t given specific instructions to report honestly. In both cases, vague direction plus outcome pressure made dishonesty more likely.
Why does this happen? One of the study’s co-authors explained this dynamic as moral distance. AI can create just enough separation between a person and the action taken that accountability begins to blur. It becomes easier to rationalize a questionable result when the machine helps produce it. In other words, they blame it on AI. Given that the accounting profession is built on trust, that should get your attention.
If AI can influence not only output but also human behavior, then an ethical culture matters more than ever.
Culture is one of those terms that gets used often but defined loosely. My working definition is that culture consists of purpose, values, beliefs, priorities, behaviors, norms, and relationships that shape the shared experience of an organization’s people. Simply put, culture is “how we do things around here,” especially when no one is watching.
That definition matters because it points to a key feature: AI can’t replace culture, but it can amplify or diminish whatever culture is already in place. For instance, in an organization where truthfulness, accountability, and careful review are already strong, AI can become a valuable assistant. In an organization where speed, appearances, or outcomes matter more than integrity, AI can magnify those tendencies at scale. That’s why leaders can’t afford to let their ethical culture evolve by default in the AI era—they must build it by design.
Here are five steps for building an ethical culture in the age of AI:
Because AI can either amplify or diminish whatever culture is already in place, the first step to building an ethical culture should be to revisit the organizational foundation. Vision, purpose, mission, and values should do more than decorate a wall or appear on the website. They should be reviewed and perhaps updated.
If a company says it values integrity, stewardship, trust, or responsible judgment, those words should shape how AI is used. They should influence what type of work is appropriate for AI support, where heightened review is needed, and where human judgment must remain firmly in charge.
Not every AI use case carries the same ethical risk. Leaders should identify the hot spots where AI could distort truth, reduce care, or increase temptation. Those areas may include financial reporting, treasury and cash management, tax positions, audit and assurance activity, risk management, and work involving employee or client data. The key questions teams should be asking are:
With AI, risk assessment isn’t just about compliance and cybersecurity. It’s also about temptation, pressure, and moral distance.
Values matter, but values alone aren’t enough. Plenty of organizations have had impressive-sounding values and still failed to live up to them. Look no further than Enron: At the time of the company’s massive fraud scandal and collapse in late 2001, their stated values were “communication, respect, integrity, and excellence.”
Enron’s failure highlights the real challenge: operationalizing values into clear behavioral standards. Therefore, organizations should ask themselves:
Teams shouldn’t have to guess the answers to these questions. When AI is involved, leaders need to define what acceptable and unacceptable behaviors actually look like.
This is where ethical culture becomes visible. Strong guardrails help protect integrity, judgment, and trust while still allowing people to benefit from AI. To create these guardrails, begin with:
Two of the above guardrails deserve special emphasis. The first is “verifying AI-generated facts, assumptions, and conclusions.” AI can sound polished and confident while still being wrong. Output needs scrutiny, not admiration. The second is “slowing down under pressure.” One of AI’s greatest benefits is speed, but speed can become a hazard when it reduces review, increases cognitive fatigue, and encourages rubber-stamping. Ethical lapses often begin not with bad intent but with haste and overconfidence.
Finally, leaders must reinforce the culture they say they want. Make it safe for people to raise concerns when something feels off. Encourage questions, challenge assumptions, celebrate examples of good judgment, not just fast output, and when lapses occur, address them. After all, culture is shaped not only by what leaders praise but also by what they tolerate.
AI will continue to evolve and so will the expectations surrounding it. But one principle should remain steady: Technology may assist the work, but it must not weaken the integrity behind the work. In the age of AI, ethical culture isn’t a side issue—it’s one of the clearest tests of leadership.
