Employees use personal AI accounts for work, study

Harmonic Security has published research showing that employees use personal AI accounts for work 64.5% of the time. It also found that 45.6% of personal AI activity takes place on enterprise plans paid for by employers.

The findings are based on an analysis of 1,935,247 classified AI-session minutes across ChatGPT, Claude, Google Gemini, Microsoft Copilot, Perplexity and DeepSeek. The results suggest that the line between work and personal AI use does not reflect how staff actually use the tools.

The data shows a mixed pattern: employees carry out business tasks on personal or free accounts, while also using employer-funded enterprise accounts for personal activity. Overall, 74.6% of AI use at work had a clear business purpose.

Across the six platforms, task types were distributed in similar ways regardless of which tool workers used. Harmonic found that 47% of AI time went on efficiency and automation, 20% on decision support, 20% on risk and compliance, 7% on revenue and growth, and 6% on innovation.

This consistency suggests employees are not choosing one model for one type of task and another for a different one. Instead, workers often use whichever service is already open.

Department split

AI use was not evenly spread across company functions. Legal and Governance accounted for 19.5% of all AI hours in the study, making it the heaviest user group, with 81% of that use taking place on enterprise plans.

Commercial and operational teams showed a different pattern. Go-to-market teams represented 17.5% of AI minutes, but only 39% of that activity ran through enterprise plans, while Operations recorded just 18% of its AI use on enterprise accounts.

This leaves a large share of work by sales, marketing and operations staff outside the view of company dashboards and internal controls. Those teams were using AI for tasks such as proposals and competitive research on accounts employers could not see.

The report also highlights what happens when staff leave. Business context embedded in personal AI histories can leave with departing employees, including material related to contracts, strategies and deals held in accounts the employer does not own and cannot recover.

Session length

The research also examined session duration as a measure of how deeply employees engage with a tool. Claude stood out for more analytical work, with average sessions lasting 10 minutes and 12 seconds, compared with 5 minutes and 53 seconds for ChatGPT.

Harmonic argued that time spent in a session may be more useful to security teams than raw query counts. A lengthy contract review may involve far more pasted information and business context than a brief question, even though both would appear as a single event on a standard usage dashboard.

The study covered a trailing seven-week period ending in April and classified each AI conversation as personal, business or ambiguous using a large language model trained on enterprise AI usage patterns. Plan categories were derived from account metadata provided through enterprise reporting interfaces from the platforms covered in the analysis.

The figures come as companies continue to increase spending on AI software while trying to determine whether those tools are being used through approved channels. The findings suggest that buying corporate licences does not necessarily give employers a full picture of how staff use generative AI day to day.

Alastair Paterson, Chief Executive Officer and Co-founder of Harmonic Security, said the research was intended to show how behaviour looks across both personal and enterprise accounts. "Every organization is pouring money into AI right now, and almost none of them know what their people are actually doing with it. This is the first cross-platform analysis of AI use cases at scale, across personal and enterprise accounts together. It is the first genuine look at how AI is actually being used at work," Paterson said.