One-quarter of executives report that internally generated AI errors have reached the public or their boards, according to a new survey.

A poll of more than 2,000 executives around the world reveals that 26% say internal audits have shown AI mistakes. Meanwhile, only 39% could say they are “very confident” in the accuracy of AI-generated material appearing in the next annual report or board presentation.

An additional 45% said they could only say they would be “somewhat confident”.

The survey, from cloud company Workiva, captures concerns and divided views on the application of AI in business administration.

Elsewhere, the survey reveals only 11% of executives believe their data is robust enough for use in AI. More than 50% of investors are monitoring AI accuracy.

Workiva’s chief financial officer, Barbara Larson, says: “Confidence in AI without control over data quality is a liability, not a strategy.” She adds: “Getting this right is about more than avoiding errors. Business leaders can move faster and embed AI deeper into their operations when they trust what their systems produce. That’s a real competitive edge.”

Embarrassing errors

The survey comes following a run of embarrassing stories about companies issuing reports tainted by AI errors. In July, it was reported that PwC had published four “thought-leadership” reports containing unverifiable information.

In June, KPMG also withdrew a report that contained AI hallucinations.

There are long-standing warnings about the reliability of AI-generated material, and concerns have reached boardrooms and other functions. Earlier this year, the Oxford-Globescan Corporate Affairs survey revealed that 43% of those polled globally said they lacked preparation for a crisis caused by AI misinformation. The same survey found that AI was viewed as a risk second only to geopolitical uncertainty.

Research by watchdogs at the Financial Reporting Council (FRC) earlier this year showed that AI is making its way into corporate reporting in the UK, but executives appear cautious. Most of those using it say it is for “narrative reporting” and few appear to be deploying AI for financial reports. That said, 23% of those polled say they use AI for calculations and 24% for “materiality assessments”.

At the time, the FRC said in a statement: “While companies are beginning to use AI and other technologies to support reporting processes, most deployment is focused on lower-risk, task-specific activities rather than areas requiring significant professional judgement.”