AI Needs More Human Accountability

Imagine an employee using artificial intelligence (AI) to prepare an important report. The system produces a well-written document within minutes, complete with convincing explanations and supporting information. The employee makes a few adjustments and submits it to management.

Later, someone discovers that several statements are inaccurate. A reference does not exist, and one recommendation is based on incorrect information.

Who should take responsibility? The employee who submitted the report, the manager who approved it, or the AI system that generated the information?

This situation illustrates a growing challenge in modern workplaces. AI is making many tasks easier and faster, but it is also changing how people make decisions and take responsibility for their work.

The question is no longer simply what AI can do. It is also how we remain accountable when using it.

AI Is Useful, but Not Infallible

There are good reasons why organisations are adopting AI. It can summarise documents, analyse information, draft correspondence, assist customer service teams, and support decision-making.

Research by Erik Brynjolfsson and colleagues, published in The Quarterly Journal of Economics, examined AI assistance among 5,172 customer support agents. The study found that access to AI increased productivity by approximately 15%, with particularly meaningful benefits for less experienced workers.

Such findings demonstrate why organisations are interested in these technologies. When used appropriately, AI can improve productivity and help employees perform tasks more efficiently.

However, faster work is not necessarily better work.

AI-generated information may appear convincing even when it contains errors. A well-written explanation does not guarantee factual accuracy, and a confident recommendation does not necessarily reflect good judgement.

This becomes particularly important when AI is used in activities involving financial decisions, public services, recruitment, healthcare, or institutional communication.

In these situations, the consequences of mistakes may extend beyond the person using the technology.

When AI Makes Mistakes, Who Answers?

In 2024, a case involving Air Canada attracted international attention after its website chatbot provided incorrect information about discounted fares for passengers travelling because of a family bereavement.

A customer relied on the chatbot’s explanation but was subsequently denied the expected refund.

The British Columbia Civil Resolution Tribunal found Air Canada responsible for the misleading information and awarded compensation. The case, discussed by the American Bar Association, illustrates an important principle: organisations cannot necessarily avoid responsibility simply because incorrect information originated from an automated system.

Although the ruling concerned a specific dispute, its management implications are much broader.

When an organisation introduces technology into its services, it must also consider how the technology will be monitored, how errors will be corrected, and who will respond when something goes wrong.

The technology may generate the answer, but the organisation remains responsible for managing the service.

The Danger of Trusting Without Checking

One concern with AI is that people may gradually become less critical of the information it produces.

When a system repeatedly delivers useful results, users may begin assuming that its future outputs are equally reliable. Over time, checking becomes less frequent, especially when employees are under pressure to complete tasks quickly.

This creates a problem when AI-generated material is treated as verified information rather than a starting point for further examination.

According to the Stanford AI Index 2026, the number of documented AI-related incidents increased from 233 in 2024 to 362 in 2025. These figures reflect reported incidents rather than every problem that occurs, but they demonstrate growing concerns surrounding AI reliability and governance.

For organisations, the lesson should not be to discourage employees from using AI. Instead, employees need to understand when its outputs require verification.

A draft email may need a simple review. A financial analysis may require checking the underlying calculations and assumptions. A recommendation affecting someone’s employment or access to services deserves considerably greater scrutiny.

Different tasks involve different levels of risk. The responsibility for verification should reflect those differences.

Human Oversight Must Be Meaningful

Many organisations respond to AI-related risks by introducing a policy requiring human review before information is released or decisions are approved.

This is a reasonable starting point, but it may not be sufficient.

Consider an employee reviewing a detailed AI-generated report without having the knowledge, time, or access to information needed to assess its accuracy. The employee may approve the report simply because it appears professionally written.

Technically, human review has taken place. In practice, very little has been verified.

The US National Institute of Standards and Technology, through its AI Risk Management Framework, emphasises the importance of clearly defined responsibilities, risk management, and trustworthy AI practices.

Effective oversight therefore requires more than asking someone to approve an AI-generated output.

Employees must understand the limitations of the tools they use. Managers should provide appropriate guidance, particularly when AI is introduced into important organisational processes.

There must also be room for employees to question AI-generated recommendations rather than assuming that disagreement with the system reflects a lack of technological understanding.

Accountability Begins with Management

Responsibility for AI should not rest entirely on individual employees.

Organisations decide which technologies to adopt, what information can be processed, and which tasks may be assisted or automated. These decisions influence how employees eventually use AI.

Managers therefore have a responsibility to establish reasonable expectations.

Before implementing AI, organisations should identify its intended purpose, potential risks, and appropriate safeguards. During implementation, they should monitor how the technology performs and address problems as they emerge.

The OECD AI Principles emphasise accountability, transparency, and responsible risk management throughout the AI system’s lifecycle.

Similarly, UNESCO’s Recommendation on the Ethics of Artificial Intelligence stresses that human responsibility and accountability should not be displaced by AI systems.

These principles are relevant beyond large technology companies. Schools, universities, small businesses, charities, and public institutions increasingly use AI in everyday activities.

They may not possess sophisticated AI governance departments, but they can still establish clear procedures, provide basic training, and encourage responsible use.

Technology Changes, Responsibility Remains

AI will continue to influence how organisations operate. Some tasks will become easier, while others will require employees to develop new skills and exercise greater judgement.

The objective should not be to resist technology or insist that humans perform every task manually.

Instead, we should recognise that adopting AI changes the nature of professional responsibility.

Employees may spend less time preparing initial drafts but more time checking information, interpreting results, and considering consequences. Managers may gain faster access to analysis, but they must still decide whether the recommendations are appropriate.

Ultimately, accountability cannot be transferred to a machine simply because the machine helped produce an outcome.

The more organisations depend on AI, the more important it becomes to ensure that people remain capable of questioning, correcting, and taking responsibility for its results.

The views expressed in this article are solely those of the author and do not necessarily reflect the views of The Opinion Desk.