Before AI goes wrong, business leaders should identify harmful outcomes and preserve human judgment, a Business Record leadership column argues. Contributors warn against accepting confident answers without questioning their accuracy.
The column connects an incorrect travel chatbot answer with larger organisational risks. Its focus is prevention through specific responsibilities, verification and opportunities for employees to develop their own skills.
Before AI Goes Wrong, Define The Harm
The writer describes receiving a polished, plausible answer while planning travel. Checking another source revealed a timing mistake that could have disrupted the trip.
That experience prompted questions about errors affecting other people or higher-stakes decisions. A convincing answer, the column argues, is not necessarily accurate.
Speed can also encourage users to skip judgment and verification. The article discusses AI in research, writing, information analysis, customer communications and decisions.
In the Business Record column, AI risks extend beyond unauthorised access to systems. Examples include inaccurate information, biased recommendations, confidential information exposure and fabricated sources.
The column also identifies actions taken with insufficient human scrutiny. It cautions leaders against treating AI risk and cybersecurity as identical problems, even when they overlap.
Related coverage examines a separate AI cybersecurity test. That report concerns access to company systems rather than the leadership contributors’ advice.
Work Backward From A Specific Failure
The column draws on Reid Blackman’s May 2026 Harvard Business Review article, “What Are Your Company’s AI Nightmares?” It presents his recommendation to identify unwanted outcomes and work backward.
His criticism concerns policy efforts that become too slow, vague or difficult to communicate. Specific failure scenarios give leaders a more concrete starting point alongside broader principles and policies.
Examples include a hiring tool excluding qualified candidates or an employee sharing confidential information with a public platform. Another scenario involves a chatbot giving customers inaccurate or harmful advice.
AI-generated misinformation could also enter an executive presentation and influence a major decision. These are risks described by the column, rather than reported incidents at the contributors’ organisations.
Before AI goes into important decisions, the column urges clarity about who prevents failures and when human review is required. Employees also need the judgment and tools to recognise problems before they spread.
Separate related reading considers AI data control and technology investment outcomes. Those articles address different questions about organisational technology.
Human Review Before AI Goes Into Decisions
Samantha Mosser, president and CEO of Bankers Trust Company, warns against removing human judgment from important decisions. She says AI should support decision-making, particularly where clients, employees, risk or strategy are affected.
Mosser also calls for regular audits and validation of AI outputs. Her concern is people ceasing to question recommendations, rather than simply acknowledging that a system can make mistakes.
Dave Leto, CEO of Palmer Group, emphasises relationships and human potential. He cautions against automation-centred growth strategies that neglect the people supporting long-term success.
Skylar Mayberry-Mayes, executive director of Grand View University, Jacobson Institute, raises another concern. AI efficiency should not come at the expense of opportunities for people to develop capabilities.
Mayberry-Mayes also serves as vice chair of the Des Moines School Board. The contribution emphasises preserving experiences that help employees practise critical thinking and innovation.
The column cites leadership author Kari L. Zeller’s distinction between productivity and value. Producing more activity does not establish that the activity was worth producing, she argues.
“Productivity measures how much activity we can produce,” Zeller says. “Value measures whether the activity was worth producing.”
The column says fear should not stop organisations from experimenting and learning. It nevertheless asks leaders to judge whether outputs are accurate, useful and appropriate.
Leaders cannot anticipate every failure, the column says. Its proposed next step is identifying specific harmful outcomes, assigning responsibility and establishing verification and human review before those problems occur.
