Who Owns the Wrong Answer?

Does incorrect information a company provides about itself become any less incorrect when AI produces it faster?

On 8 October, Google announced its new Gemini agent, which can connect to enterprise systems, plan work, use different tools and deliver results.

On the same day, the European Commission fined Irish building-materials manufacturer Kingspan €40 million.

The reason was the provision of incorrect and misleading information during the merger review of its proposed acquisition of Slovenia’s Trimo. The Commission examined matters including market penetration, the availability of tender data and the board’s involvement in certain issues. Kingspan said it would appeal.

The two events are not directly connected. The Kingspan case was not an AI error.

But placing them side by side reveals an uncomfortable reality for corporate management.

Technology is increasingly preparing to speak on a company’s behalf. The law still looks to the company to stand behind what is said.

And the issue goes beyond legal responsibility.

The figure in a sales report may not equal the amount that can actually be collected. Orders, invoices, revenue and cash are different concepts. A customer contract marked active in a system does not mean future revenue from that customer is guaranteed.

In M&A, every one of these distinctions can change a company’s value.

For example, if a non-recurring item of income is treated as recurring in the normalised EBITDA presented to a buyer, the error does not only affect that period’s profit. It carries through to the company’s valuation via the transaction multiple. The same error can also affect acquisition financing, debt-service capacity and price negotiations.

AI can prepare this table in seconds.

But we cannot assume it automatically understands why the underlying records differ, the rationale for each accounting adjustment or which assumptions are defensible.

The danger is not only that the system may give an obviously wrong answer, but that it may deliver the wrong answer with extraordinary conviction.

A poorly prepared table makes people suspicious. A presentation that looks flawless can sometimes discourage scrutiny.

Where does Türkiye stand in this debate?

According to TurkStat’s research dated 2 October 2026, AI use among Turkish enterprises with at least 10 employees rose from 7.5% to 14% in one year.

There is an even more interesting finding.

Among enterprises that do not use AI but are considering it, 66.4% cite uncertainty over legal liability for harm caused by the technology as one reason for not using it. A lack of expertise is cited by 72.3%, while 65.4% point to data security and privacy concerns.

These cannot be dismissed as abstract fears of companies reluctant to embrace technology. They are concrete questions from managers who will continue to bear responsibility for the work.

Then there are financing conditions.

The Istanbul Chamber of Industry’s September manufacturing PMI stood in contraction territory at 47.9. In contrast, the CBRT’s seasonally adjusted capacity utilisation rate rose by 0.6 percentage points to 74.1% in the same month. These indicators do not measure the same thing, so it would be wrong either to say production in Türkiye has stopped altogether or to claim that the sector is now comfortable.

Following the CBRT’s September decision, the policy rate stands at 37%. The meeting summary published on 17 September puts the TRY commercial lending rate for the week of 4 September at 49.3%. Although the latest lending survey points to stronger credit demand and expectations of easing standards in the final quarter, financing costs remain an important investment filter.

In this environment, a Turkish company does not necessarily face the same question as a company in Silicon Valley.

While one is considering how many processes can become autonomous, the other may prioritise shortening collection periods, releasing working capital or identifying a mispriced risk in time.

Buying the most advanced system is therefore not always the best investment.

Particularly in financial reporting, credit assessment and acquisitions, the economic value of speed cannot be measured only in working hours saved. Preventing a poor decision, identifying a risky customer early or negotiating an acquisition price on sound data are also part of that value.

But first, the company must be able to trust its own records.

Before granting AI authority, a company must establish who is responsible for its own information.

If it is unclear which data should be used, what date it relates to, what adjustments it has undergone and which person approved it, automation merely accelerates uncertainty.

Google’s new agents promise to complete tasks. The Kingspan decision reminds companies how seriously their obligations regarding the accuracy of information are taken. This does not mean every company needs a burdensome control framework; it means controls should strengthen as financial significance and legal risk increase.

For a decision-maker, the starting point is not the process that seems easiest to automate, but the process that could cause the greatest economic damage if it goes wrong. First establish its data, its owner and its validation method. Then measure the financial return on the speed technology brings.

Research sources

The main developments and indicators in this article were verified against the following sources: Google Cloud — Gemini Agent announcement, 8 October 2026; European Commission — Kingspan decision, 8 October 2026; TurkStat — Artificial Intelligence Statistics, 2 October 2026; Istanbul Chamber of Industry — September 2026 Manufacturing PMI; CBRT — September 2026 capacity utilisation.