The CEO Can’t Blame the Algorithm

The following article was originally published on Strategic, the platform for communication leadership globally, and is being republished with permission.

Every generation of senior management encounters a technology it is tempted to treat as somebody else’s responsibility.

Today, that technology is artificial intelligence.

The danger is not simply that executives may misunderstand AI. It is that they may authorize decisions they cannot explain, approve claims they cannot substantiate and discover, too late, that the public holds management accountable for machines operating in the company’s name.

The CEO owns the outcome, but the CCO owns the message, and both answers start with the same five questions.

1. Where is AI already speaking or acting on behalf of the company?

Senior management must first understand where AI is being used. It may be answering customer questions, screening job applicants, preparing financial analyses, evaluating employee performance, recommending products, monitoring operations or drafting public statements. In some companies, employees may be using publicly available AI tools without formal authorization or oversight.

Management cannot govern what it has not identified. The company therefore needs an inventory of every significant AI application, including who authorized it, what information it uses, which stakeholders it affects and whether a person reviews its output before action is taken. The greater the potential consequence, the greater the need for human oversight.

2. Who verifies the information and recommendations AI produces?

AI can generate information that is convincing, polished and wrong. It can reproduce inaccuracies, overlook context and reflect biases embedded in its training data. Fluency must never be mistaken for accuracy.

Every important AI application should have an accountable human owner. That person must be qualified to question the output, verify important facts and stop its use when the results cannot be trusted. Review cannot be treated as a ceremonial approval at the end of the process. It must be an active exercise of professional judgment.

The critical question is not simply whether a person participated. It is whether that person had the knowledge, authority and time to challenge what the technology produced.

Consider a company whose AI-driven customer service tool began offering refund terms that exceeded policy because it had learned, from thousands of past chats, that generous answers closed tickets faster. No one instructed it to overpromise. No one caught it until customers started asking why the fine print didn’t match what the bot had told them. The tool wasn’t malicious. It was optimizing for the wrong thing, unsupervised, and it did so in the company’s name.

3. What must the company disclose to employees, customers and investors?

Stakeholders should not have to discover on their own that AI influenced a decision affecting them. Employees may deserve to know when AI is used in hiring, evaluation or workforce planning. Customers may need to know when they are interacting with a machine or when an automated system is influencing the products, prices or services offered to them. Investors need a candid explanation of how AI affects the company’s strategy, operations and risks.

Disclosure does not require revealing proprietary technology. It requires explaining AI’s role honestly and in language stakeholders can understand. Management should determine what people reasonably need to know, when they need to know it and how the company will respond to questions or objections.

Transparency is not merely a legal exercise. It is part of the company’s obligation to maintain credible relationships with the people affected by its decisions.

4. Who accepts responsibility when AI causes harm?

Responsibility cannot be assigned to an algorithm. Nor should management be permitted to shift blame to a software provider, consultant or employee after a problem becomes public. If the company authorized the technology, benefited from it and acted on its recommendations, the company owns the consequences.

Before deploying an AI system, management should decide who has the authority to approve it, monitor it, suspend it and respond when it fails. The company also needs a process for investigating mistakes, correcting harmful decisions and supporting the people affected.

When the lawsuit or the headline comes, who signs their name to the correction? Stakeholders will want to know who made the decision and what the company intends to do about it. “The algorithm did it” will not be an acceptable answer.

5. Can the CEO explain the company’s use of AI clearly, without hiding behind technical language?

A CEO does not need to be a computer scientist. But the CEO must understand enough to explain where AI is being used, why the company is using it, what risks it creates and how those risks are being managed.

Technical language often becomes a shield against accountability. References to models, data sets and automated processes may describe how a system operates, but they do not answer the questions stakeholders care about: Was the decision fair? Was the information accurate? Was anyone harmed? Who was responsible?

If the CEO cannot explain the company’s use of AI in plain language, senior management may not understand it well enough to govern it. The communications executive has an important role here, not simply translating technical terminology, but ensuring that management confronts the ethical, operational and reputational consequences of its decisions.

Senior management may delegate the operation of artificial intelligence, but it cannot delegate responsibility for what the technology says, decides, conceals or gets wrong.

Brand matters. Don’t leave it to AI to solve the issues.