Does your company use artificial intelligence? Are you sure it knows the law as well?

In the past year I have noticed one curious thing.

Companies no longer debate whether they will use artificial intelligence. The discussion has moved much further ahead. They are now seeking licences, testing platforms, comparing models, drafting documents, translating contracts, replying to e-mails and generating reports at a pace that, until recently, we associated exclusively with human work.

The change has happened so quickly that many organisations have not even realised when they crossed from the realm of experiments into the realm of business processes.

And it is precisely here that the question arises which, in my view, is almost entirely missing from every discussion about artificial intelligence.

What, exactly, have you bought?

Most directors will answer without hesitation:

“A tool that helps us work faster.”

The answer is not wrong.

It is, however, incomplete.

The moment an artificial-intelligence system drafts contractual clauses, analyses CVs, formulates replies to customers, prioritises requests, proposes solutions or generates content that will influence a business decision, the company is no longer merely using software. It has introduced into its own processes a new participant that produces results, influences people and generates legal consequences.

That is where the discussion begins which many organisations still avoid.

Who is liable when the artificial intelligence gets it wrong?

The question seems simple, yet the answer is less comfortable than most managers would wish.

The artificial intelligence itself does not answer.

The company will answer.

The director will answer if his or her supervisory duties have been breached.

The employee will answer who used the system’s output without due care.

In certain situations the technology supplier will answer.

Rarely, however, will the system itself answer.

I observe that many companies discuss artificial intelligence almost exclusively through the lens of efficiency. How many hours do we save? How many staff can we redeploy? How quickly can we draft a contract? How quickly can we reply to a customer?

These are legitimate questions.

There is, however, another category of questions that land on the legal counsel’s desk and that, as a rule, appear far too late.

Who checks the result produced by the AI before it takes effect?

In what situations is the employee obliged to intervene?

How do we demonstrate that the final decision was made by a human and not by an algorithm?

What information may be fed into an artificial-intelligence model without breaching confidentiality obligations?

What do we do if the system generates discriminatory, erroneous or simply false content?

These questions are not technical.

They are governance questions.

For this reason I believe many organisations make a mistake from the very first moment. They treat the implementation of artificial intelligence as an IT project. In reality, it is a management project. The technical department installs the application. Management decides where it will be used. The legal counsel is the one who must set the rules under which it may operate without turning today’s efficiency into tomorrow’s litigation.

In recent months I have repeatedly heard the same question:

“Does the law allow us to use AI?”

I believe the correct question is different.

How do we use AI so that the final decision remains a human one and the legal responsibility can be explained and defended?

The difference between the two questions is essential.

The first seeks approval.

The second builds a process.

And the difference between a company that merely buys technology and one that knows how to integrate it responsibly will not be determined by the artificial-intelligence model it uses.

It will be determined by the people who have understood that any new technology, above all, changes the rules by which the organisation assumes its decisions.