Kolmogorov survey flags AI risks

1/ Kolmogorov Law said September 2 that a survey of 500 employed U.S. adults who use AI at work found many workers are acting on answers they do not fully trust. The survey was fielded on September 1 through Pollfish and focused on how often workers verify AI output, what happens when they do not, and who they think is legally responsible when errors cause harm. (prnewswire.com) 2/ The headline numbers were direct. Kolmogorov Law said 65% of workers using AI at work do not always verify its answers before acting on them. (prnewswire.com) It also said 42% had knowingly accepted an answer they suspected was wrong, 30% said a wrong AI answer had already caused a problem at work, and only 22% said their employer had a written rule requiring verification. 3/ The survey also found 51% of respondents believed they would be personally legally responsible if they acted on a wrong AI answer. That matters because the data suggests many workers understand the liability question but still skip routine checking. Kolmogorov Law framed that gap as a workplace compliance problem, not just a productivity issue. (prnewswire.com) 4/ The most telling number may be the 22% with a written verification rule. If that figure is accurate, most AI users in the survey are operating without a formal employer requirement to confirm outputs before using them in work. That leaves verification as an individual habit rather than a documented process. (prnewswire.com) 5/ The 30% who reported a workplace problem tied to a wrong AI answer is the number to watch operationally. A survey result like that does not tell us how severe those problems were, or in which industries they occurred, but it does indicate that bad outputs are not a theoretical risk inside offices already using these tools. (prnewswire.com) 6/ The legal angle is central because Kolmogorov Law is a law firm, and the release is written from that perspective. The firm’s framing is that workers may wrongly assume AI shifts accountability away from the human user. Its finding that half of respondents think they would personally be liable suggests at least some workers already understand that the person who acts on the output can still bear responsibility. (prnewswire.com) 7/ The broader workplace context points in the same direction. SHRM said in its 2026 workplace AI research, based on 5,875 U.S.-based workers surveyed in March and April, that AI use is already widespread across work and personal tasks. That makes verification rules more important as AI moves from occasional experimentation into routine workflow. 8/ In practical terms, the survey is a warning about process discipline. (prnewswire.com) The risk is not only that AI can be wrong; it is that workers may use confident-sounding output without a required check, especially when speed is rewarded. In settings such as clinical documentation, decision support, finance, legal review, or compliance work, that kind of failure point can be built into the workflow unless employers make verification explicit. The Kolmogorov Law release did not break out healthcare separately, but its numbers describe the kind of conditions under which those risks can spread. 9/ The next concrete step is publication and scrutiny. Kolmogorov Law has posted the survey findings on its website, and the release says the sample covered 500 employed U.S. adults who use AI on the job, fielded September 1, 2026, through Pollfish. (prnewswire.com) (shrm.org)