The Proportion of Entering "Confidential Information" into Shadow AI is About Twice as High for "Manager/Director Class" Compared to General Employees
According to a survey by 'Hoshi no Manabi Cafe' (GRAS Group Inc.), managers are about twice as likely (37.5%) to input confidential information into unauthorized 'shadow AI' compared to general employees (18.8%), highlighting severe enterprise security risks.
📋 Article Processing Timeline
- 📰 Published: April 14, 2026 at 23:50
- 🔍 Collected: April 14, 2026 at 15:31
- 🤖 AI Analyzed: April 19, 2026 at 15:42 (120h 10m after Collected)
GRAS Group Inc., operating the career support media for women, 'Hoshi no Manabi Cafe', conducted an 'Anonymous Survey on the Actual Usage of Generative AI' targeting 478 company employees who use generative AI in their business operations.
■ Background of the Survey
While the rapid spread of generative AI is improving operational efficiency in many companies, there are growing concerns about the risk of 'Shadow AI'—the use of personal generative AI accounts for business purposes that are not authorized or monitored by the company. This survey was conducted to highlight the unvarnished 'realities of AI usage in the field' and 'hidden security risks', aiming to clarify the true measures companies should take.
■ Topics
[Basic Reality] The most common uses overall are 'Idea generation (57%)' and 'Email drafting (46%)'. Integration into daily work is progressing.
[Core of the Risk] The proportion of entering 'customer data' and 'contracts' into Shadow AI is about twice as high for manager/director class (37%) compared to general employees (18%).
[Limits of Rules] Regardless of official or unofficial use, about 40% of managers input confidential information into AI.
[Survey Overview]
Survey Name: Anonymous Survey on the Actual Usage of Generative AI
Survey Period: April 2026
Survey Target: Company employees using generative AI for business
Valid Responses: 478 (251 Shadow AI users, 227 users of authorized AI only*)
Survey Method: Internet survey (anonymous)
*Note: In this survey, 'Shadow AI users' and 'Authorized AI only users' were recruited separately with specific conditions. Therefore, the breakdown of valid responses above does not indicate the overall usage rate (incidence rate) of shadow AI among all generative AI users in society.
[Regarding Citation]
When citing data, images, or text from this article, please ensure to include a link and state that it is cited from the original article, 'Hoshi no Manabi Cafe'.
(Example) Source: Hoshi no Manabi Cafe 'The Proportion of Entering "Confidential Information" into Shadow AI is About Twice as High for "Manager/Director Class" Compared to General Employees'
1. [Basic Reality] Generative AI Penetrating Daily Work. Top Uses are 'Idea Generation' and 'Emails'
When we asked all respondents using generative AI for business 'for what purposes they use it (multiple selections)', the top answer was 'Idea generation for planning and naming (57.1%)', followed by 'Drafting emails for external communication (46.4%)' and 'Creating reports such as daily reports and training reports (35.5%)'. These results suggest that generative AI has become widely established for streamlining daily text-based tasks.
2. 'About 1 in 4' Shadow AI Users Input Confidential Information
However, behind this integration into daily work, serious security risks are emerging. When we investigated what specific data was being input by 'Shadow AI users (251 people)'—those using personal AI accounts without reporting to the company—we found that 23.1% of the total (about 1 in 4) were inputting highly sensitive information such as 'customer lists and sales results' and 'important documents like contracts'.
3. More Dangerous than Junior Staff? The Fact that 'About 40%' of Managers Input Confidential Information
Furthermore, when we analyzed this reality by job title, an unexpected fact emerged. While the percentage of general employee class inputting confidential information was 18.8%, it jumped to '37.5% (about double)' among middle management in the manager/director class. The data showed a stronger tendency among managers who likely have decision-making authority and the rights to handle important data.
Hypotheses for this include that they have more opportunities to handle important data in the first place, or that they feel less sense of crisis regarding company rules compared to general employees, but further investigation seems necessary on this point.
4. There is an Inherent Need to Have AI Process Important Information
So, will simply tightening rules by stating 'Inputting confidential information into AI is prohibited' solve the shadow AI problem?
To unravel this answer, this survey deliberately conducted a comparative analysis of the usage reality among 'the demographic provided with secure, authorized AI by their company'. As a result, it was found that 40.5% of managers using authorized AI input confidential information into AI as part of their business operations.
Of course, if it is an 'authorized AI' for corporate use where input data is not used for learning, and the company permits it
■ Background of the Survey
While the rapid spread of generative AI is improving operational efficiency in many companies, there are growing concerns about the risk of 'Shadow AI'—the use of personal generative AI accounts for business purposes that are not authorized or monitored by the company. This survey was conducted to highlight the unvarnished 'realities of AI usage in the field' and 'hidden security risks', aiming to clarify the true measures companies should take.
■ Topics
[Basic Reality] The most common uses overall are 'Idea generation (57%)' and 'Email drafting (46%)'. Integration into daily work is progressing.
[Core of the Risk] The proportion of entering 'customer data' and 'contracts' into Shadow AI is about twice as high for manager/director class (37%) compared to general employees (18%).
[Limits of Rules] Regardless of official or unofficial use, about 40% of managers input confidential information into AI.
[Survey Overview]
Survey Name: Anonymous Survey on the Actual Usage of Generative AI
Survey Period: April 2026
Survey Target: Company employees using generative AI for business
Valid Responses: 478 (251 Shadow AI users, 227 users of authorized AI only*)
Survey Method: Internet survey (anonymous)
*Note: In this survey, 'Shadow AI users' and 'Authorized AI only users' were recruited separately with specific conditions. Therefore, the breakdown of valid responses above does not indicate the overall usage rate (incidence rate) of shadow AI among all generative AI users in society.
[Regarding Citation]
When citing data, images, or text from this article, please ensure to include a link and state that it is cited from the original article, 'Hoshi no Manabi Cafe'.
(Example) Source: Hoshi no Manabi Cafe 'The Proportion of Entering "Confidential Information" into Shadow AI is About Twice as High for "Manager/Director Class" Compared to General Employees'
1. [Basic Reality] Generative AI Penetrating Daily Work. Top Uses are 'Idea Generation' and 'Emails'
When we asked all respondents using generative AI for business 'for what purposes they use it (multiple selections)', the top answer was 'Idea generation for planning and naming (57.1%)', followed by 'Drafting emails for external communication (46.4%)' and 'Creating reports such as daily reports and training reports (35.5%)'. These results suggest that generative AI has become widely established for streamlining daily text-based tasks.
2. 'About 1 in 4' Shadow AI Users Input Confidential Information
However, behind this integration into daily work, serious security risks are emerging. When we investigated what specific data was being input by 'Shadow AI users (251 people)'—those using personal AI accounts without reporting to the company—we found that 23.1% of the total (about 1 in 4) were inputting highly sensitive information such as 'customer lists and sales results' and 'important documents like contracts'.
3. More Dangerous than Junior Staff? The Fact that 'About 40%' of Managers Input Confidential Information
Furthermore, when we analyzed this reality by job title, an unexpected fact emerged. While the percentage of general employee class inputting confidential information was 18.8%, it jumped to '37.5% (about double)' among middle management in the manager/director class. The data showed a stronger tendency among managers who likely have decision-making authority and the rights to handle important data.
Hypotheses for this include that they have more opportunities to handle important data in the first place, or that they feel less sense of crisis regarding company rules compared to general employees, but further investigation seems necessary on this point.
4. There is an Inherent Need to Have AI Process Important Information
So, will simply tightening rules by stating 'Inputting confidential information into AI is prohibited' solve the shadow AI problem?
To unravel this answer, this survey deliberately conducted a comparative analysis of the usage reality among 'the demographic provided with secure, authorized AI by their company'. As a result, it was found that 40.5% of managers using authorized AI input confidential information into AI as part of their business operations.
Of course, if it is an 'authorized AI' for corporate use where input data is not used for learning, and the company permits it