Nearly 2 in 5 US workers have put company information into personal AI accounts, and most don't know it can be illegal

The confidential company document has a new address: someone's personal chatbot history. In a July 2026 survey of 500 employed U.S. adults commissioned by Kolmogorov Law, a California business-litigation firm,and conducted through the Pollfish research platform, 38% admitted they have entered at least one type of work information into a personal AI account their employer does not control. Additionally, 64.4% did not know that doing so can, in some circumstances, break the law.

The amount of sensitive company data moving into personal AI accounts is not trivial. Nearly a quarter of all workers surveyed (23%) say they have pasted internal emails, memos, or documents into a personal AI account. More than one in ten admit to entering financial or sales figures (12.4%), customer or client information (11.8%), contracts or legal documents (11.4%), or employee and HR information (10.6%); 9% have entered computer code or technical materials. Among only the workers who admitted putting anything into a personal account, internal documents lead at 60.5%.

The survey asked workers directly about behavior that company security monitoring has already suggested is happening. Monitoring firm Cyberhaven, analyzing millions of employee prompts, has reported that roughly 11% of what employees paste into chatbots is sensitive material, and that the overwhelming majority of workplace chatbot use runs through non-corporate accounts invisible to employers. The new data suggests workers themselves will confirm the pattern when asked: 57.6% use AI tools for work, and 36.8% do so at least partly through a personal account.

Workplace AI policies remain unclear

Workplace policy has not kept up. According to the Kolmogorov Law survey, only 35.8% of workers say their employer has a clear, written policy on what work information may be shared with AI tools. Forty percent say no such policy exists, and another 24.2% aren't sure — meaning nearly two-thirds of the American workforce (64.2%) is improvising. A further 6.6% say their employer bans AI tools outright, a rule past statistics suggest is not airtight.

The survey's starkest finding may be what happens when workers leave. When asked to imagine departing a job while their personal AI account contained chats with work material (drafts, analyses, perhaps customer details), 16.2% of respondents said they would keep the account as it is, on the theory that "the chats are mine." Another 31.6% admitted they had never thought about the question until the survey posed it. Just 19% would ask their employer what to do. For decades, the signature trade-secret dispute began with a departing employee and a copied customer list. The data suggests the new dispute may be a chat log that simply walks out the door by default.

Workers are uncertain about the legal risks

Whether any of this is unlawful depends on what was shared and how. Under the California Uniform Trade Secrets Act (Civil Code section 3426 et seq.) and the federal Defend Trade Secrets Act, disclosing information that derives value from secrecy can constitute misappropriation. Courts have not treated the fact that information was shared with an AI chatbot instead of a competitor as a legal defense. California Penal Code section 502 separately addresses knowingly taking, copying, or using data from an employer's computer system without permission and carries both civil remedies and criminal penalties. Yet, according to the Kolmogorov Law survey, only 35.6% of workers surveyed know that entering confidential company information into a personal AI account can be against the law. Fourteen percent believe it is at most a company policy issue, 6% see no problem at all, and a striking 44.4% simply aren't sure.

Employees expect to bear responsibility

When asked who should bear the most legal responsibility when company information leaks through a personal AI account, survey respondents pointed first at themselves: 42.8% chose the employee, 22.4% said employee, employer, and AI company should share responsibility equally, 16% chose the employer, and just 6.8% blamed the AI company. Workers, in other words, largely accept that the person doing the pasting owns the consequences — they just don't know how serious those consequences can be.

The findings suggest AI adoption in many workplaces is outpacing both formal governance and employee understanding of applicable policies and law. For employers, the numbers argue that an unwritten AI policy is now a measurable liability: Two-thirds of employees either have no rules to follow or don't know the rules exist. For workers, the numbers describe a risk most of them are carrying blind: More than a third of the workforce has already put company information somewhere the company cannot reach, and most of them believe the worst that can happen is a policy violation.

Methodology

This article is based on a survey of 500 employed U.S. adults age 18 and older, conducted July 8, 2026, through the Pollfish online research platform. Respondents were screened by Pollfish audience targeting to include only adults employed for wages (full- or part-time) or self-employed. The survey included two segmentation questions and eleven substantive questions covering workplace AI use, handling of confidential information in personal AI accounts, employer AI policies, awareness of applicable law, experiences with AI meeting notetakers, and recording-consent expectations, plus an attention-check question that 100% of the final sample answered correctly. Questions about consent in meetings where an AI notetaker was present were asked of the 167 respondents who reported such meetings; percentages for that subsample carry a wider margin of error of approximately plus or minus 7.6%. The margin of error for the full sample is approximately plus or minus 4.4% at a 95% confidence level.