Why Personal ChatGPT Usage at Work Is a Cybersecurity Concern

Artificial intelligence (AI) tools such as ChatGPT are transforming the way organizations operate. Employees use AI to summarize documents, draft emails, analyze information, generate code, and improve customer support workflows. While these capabilities can significantly increase productivity, they can also introduce cybersecurity risks when AI usage is not properly managed.

One growing concern is employees using personal ChatGPT accounts for work-related tasks, even when their organization already provides an approved enterprise AI environment.

For example, an employee may copy customer support tickets into a personal ChatGPT account to generate a response, upload an internal document for summarization, or submit proprietary source code to troubleshoot an application. These actions may seem harmless, but they can expose sensitive information to services outside the organization's approved security controls.

This behavior is closely associated with Shadow AI—the use of AI tools without appropriate organizational approval, oversight, or security governance.

For IT administrators, security teams, and business leaders, the challenge is no longer simply deciding whether employees should use AI. It is ensuring that AI can be used productively without compromising sensitive business information.

In this article, BigFish explores the risks of personal ChatGPT usage at work and outlines practical strategies organizations can implement to strengthen AI security and prevent data leakage.

What Is Shadow AI, and Why Should Organizations Care?

Shadow AI occurs when employees use AI applications that have not been approved by their organization or operate outside established security and governance policies.

Common examples include:

• Using personal ChatGPT accounts to summarize confidential business documents.

• Copying customer support tickets into public AI tools to generate responses.

• Uploading contracts, financial reports, or internal presentations to unauthorized AI services.

• Submitting proprietary source code to AI assistants for debugging.

• Using personal AI accounts when corporate authentication is inconvenient or unavailable.

These activities do not necessarily indicate malicious intent. In many cases, employees are simply trying to complete their work more efficiently.

However, when AI usage occurs outside the organization's security framework, IT teams may lose visibility into which tools employees use, what information they submit, and whether sensitive data is being handled appropriately.

Why Enterprise AI Accounts Alone May Not Be Enough

Providing employees with approved enterprise AI accounts is an important first step, but it does not automatically eliminate Shadow AI.

Employees may continue using personal accounts because corporate authentication is inconvenient, access is temporarily unavailable, or approved tools do not adequately support their workflows.

Furthermore, an enterprise subscription does not mean every category of company data can be submitted without restrictions. Organizations must still evaluate the service's data handling practices, applicable contractual terms, access controls, and internal policies.

Effective AI governance therefore requires a combination of approved tools, practical usage policies, technical safeguards, and employee awareness.

Five Cybersecurity Risks of Using Personal ChatGPT at Work

1. Exposure of Customer Data and Personal Information

One of the most immediate risks is employees submitting customer information to AI services without verifying whether the data is permitted for that purpose.

Consider a customer support employee who wants to generate a response to a customer experiencing account access problems. To save time, the employee copies an entire support ticket into a personal ChatGPT account.

The ticket may contain names, email addresses, account details, or other personal information.

Although the employee only intends to improve the response, the information has now been transmitted to an external service under an account that may fall outside the organization's approved controls.

Organizations should establish clear rules for handling personal information in AI prompts, including when data must be removed, anonymized, or processed only through specifically approved services.

2. Reduced Visibility and Limited Incident Investigation

When employees use personal AI accounts, their activities may not be fully visible through the organization's existing monitoring systems.

Security teams might detect access to an AI website without knowing whether the employee submitted public information, confidential documents, or customer records.

If a data leakage incident occurs, investigators may struggle to determine:

• What information was submitted.

• Whether files were uploaded.

• When the activity occurred.

• Whether the data was transmitted successfully.

• Whether similar activity happened previously.

This lack of visibility can complicate incident response, risk assessment, and internal investigations.

Organizations should therefore evaluate whether their security controls provide sufficient visibility into AI usage and whether additional monitoring is needed.

3. Exposure of Proprietary Business Information

Sensitive information extends beyond personal data. Employees may also submit commercially valuable information to AI tools, including:

• Proprietary source code and application architecture.

• Internal technical documentation and system configurations.

• Business strategies and product roadmaps.

• Financial forecasts and confidential reports.

• Contracts, pricing information, and trade secrets.

• Security assessments and vulnerability details.

Unauthorized disclosure of this information may affect competitive advantage, contractual obligations, or the security of business systems.

Organizations should classify their information assets and establish clear restrictions on which data can be processed by specific AI services.

4. Compliance and Data Protection Concerns

Using AI to process personal or regulated information may raise questions about lawful processing, data sharing, retention, and compliance with applicable requirements.

For organizations operating in Thailand, the Personal Data Protection Act (PDPA) may be relevant when personal data is involved. The specific obligations depend on the circumstances, the organization's role, the nature of the data, and the processing activities.

Using a personal ChatGPT account does not automatically constitute a legal violation. However, organizations should assess whether the data was processed appropriately and whether any notification, remediation, or other compliance obligations apply.

Clear AI usage policies and appropriate technical controls can help organizations manage these risks more consistently.

5. Gaps in Cybersecurity Policy Enforcement

An organization may have a policy prohibiting employees from sharing confidential information with unauthorized external services. However, a written policy alone cannot guarantee that the policy will be followed.

For example, existing security tools may detect access to ChatGPT but lack the ability to inspect prompts or identify sensitive information before transmission.

This creates a gap between policy and enforcement.

Organizations should assess whether their existing security architecture can identify unauthorized AI usage, detect sensitive data, and apply appropriate controls at the point where information is submitted.

How IT Administrators Should Respond to Personal ChatGPT Usage

When a security team discovers that an employee is using a personal ChatGPT account on a corporate device, it should follow a structured investigation process rather than immediately assuming that a data breach has occurred.

Step 1: Identify the Account and User

Determine which employee, device, and account are associated with the activity.

Relevant information may come from identity management platforms, endpoint management systems, browser activity records, and security monitoring tools.

Where possible, distinguish between the employee's corporate identity and the personal account used to access the AI service.

Investigations should follow organizational authorization requirements and privacy policies, with access limited to information relevant to the incident.

Step 2: Determine Whether Sensitive Data Was Submitted

Detecting personal ChatGPT usage does not, by itself, prove that sensitive information was exposed.

The security team should gather available evidence to establish what happened, including whether the employee submitted work-related prompts, uploaded documents, or transmitted information from internal systems.

Key questions include:

• Was customer or employee information included?

• Did the prompt contain confidential business data?

• Were files uploaded to the AI service?

• Was the information successfully transmitted, or was the action blocked?

• Is there evidence of repeated activity?

The investigation should be proportionate to the incident and avoid collecting unrelated personal information.

Step 3: Assess the Impact and Contain the Risk

If unauthorized disclosure is confirmed, the organization should assess the severity of the incident based on the type of data involved, the scope of exposure, and applicable requirements.

Potential responses include:

• Exposed credentials or tokens: Evaluate whether they must be revoked or rotated.

• Customer or personal data exposure: Engage the relevant privacy, legal, and incident response teams.

• Confidential business information exposure: Assess business impact and implement appropriate containment measures.

• Unconfirmed exposure: Preserve relevant evidence and continue investigating before making conclusions.

Organizations should not assume that submitted information was used to train an AI model or made publicly available without evidence supporting that conclusion.

Step 4: Address the Root Cause

Personal AI usage may result from more than intentional policy violations. Employees might turn to personal accounts because corporate authentication is difficult, access is unavailable, or approved tools do not meet their immediate needs.

IT teams should resolve legitimate access problems, improve the usability of approved AI services, and clearly communicate which accounts and data handling practices are permitted.

The objective is to make secure AI usage the easiest and most practical option for employees.

Step 5: Document the Incident and Monitor for Recurrence

Before closing the incident, document the relevant user, device, timeline, evidence, risk assessment, corrective actions, and responsible stakeholders.

Security teams should also evaluate whether similar activities are occurring elsewhere in the organization.

Repeated incidents may indicate that existing policies, employee training, or technical controls need improvement.

Six Strategies to Prevent Shadow AI and Data Leakage

Responding to incidents is necessary, but preventive controls can reduce the likelihood of unauthorized data exposure in the first place.

1. Establish a Clear AI Usage Policy

Define which AI services employees may use, what purposes are permitted, and which categories of information can be submitted.

A practical policy should explain both prohibited activities and approved alternatives. Employees should know what to do when they need AI assistance but are unsure whether their data can be shared.

2. Strengthen Identity and Access Management

Implement appropriate authentication and access controls for enterprise AI services.

Organizations should manage access according to business requirements, employee roles, and the principle of least privilege.

Centralized identity management can also improve visibility into account usage and reduce reliance on personal accounts.

3. Implement Data Loss Prevention (DLP)

Data Loss Prevention (DLP) technologies help organizations identify and control the movement of sensitive information according to defined policies.

Depending on the solution and its configuration, DLP controls may identify patterns associated with personal information, financial records, credentials, or other protected data.

For AI-related workflows, organizations should evaluate controls that can inspect prompts, file uploads, or other data transfers before information leaves an approved environment.

The effectiveness of these controls depends on the applications, channels, and data types they support.

4. Classify Data Before AI Processing

Establish a data classification framework that distinguishes between public, internal, confidential, and restricted information.

Each classification should have corresponding rules for AI usage.

For example, public information may be permitted in approved AI tools, while confidential information may require additional safeguards. Highly restricted information may be prohibited from submission to certain external services altogether.

Clear classification rules help employees make appropriate decisions and enable security teams to apply more consistent controls.

5. Improve AI Usage Visibility and Security Monitoring

Organizations should understand which AI services are being accessed, who is using them, and whether their usage complies with internal policies.

Depending on the environment, security teams may use identity logs, endpoint monitoring, secure web gateways, browser controls, or specialized AI security solutions to improve visibility.

Importantly, detecting access to an AI website is not the same as inspecting the content of a prompt. Organizations should evaluate whether their existing tools provide the level of visibility and enforcement their risk profile requires.

6. Strengthen AI Security Awareness

Employees need practical guidance on how to use AI safely.

Training should cover real-world scenarios, such as submitting customer tickets, uploading internal documents, sharing source code, and handling personal information.

Employees should also understand how to proceed when approved AI services are unavailable and where to seek guidance when data handling requirements are unclear.

A combination of awareness, usable approved tools, and technical enforcement is generally more effective than relying on prohibitions alone.

Warn, Redact, and Block: Choosing the Right Prompt Security Controls

Organizations evaluating AI security solutions should consider how they can prevent sensitive information from being submitted in the first place.

Depending on the available technology, three common control approaches are:

The appropriate action depends on the organization's data classification rules and risk tolerance.

Warnings may be sufficient for some lower-risk situations, while redaction can help preserve the usefulness of a prompt without transmitting selected sensitive values. Blocking may be appropriate when the data must not leave the organization's approved environment.

Organizations should also verify when the control takes effect. A notification provided after data has already been transmitted does not offer the same preventive protection as a control that stops the transmission beforehand.

Building a More Secure AI Environment with BigFishTec

AI can improve productivity, accelerate workflows, and help organizations deliver better business outcomes. However, these benefits must be balanced with appropriate security and governance.

Preventing Shadow AI requires more than blocking access to popular AI websites. Organizations need to understand how employees use AI, which information is being processed, and where existing security controls may have gaps.

A comprehensive approach can combine AI governance, Identity and Access Management (IAM), Data Loss Prevention (DLP), security monitoring, and employee awareness to reduce the risk of unauthorized data exposure.

Organizations should begin by assessing their current AI usage, identifying sensitive data flows, and reviewing whether their existing cybersecurity controls can enforce the required policies.

BigFishTec supports organizations in strengthening their cybersecurity posture and managing evolving digital risks. By aligning security controls with business requirements, organizations can work toward adopting AI more confidently while maintaining appropriate protection for sensitive information.

Key Takeaways

• Personal ChatGPT usage at work can introduce Shadow AI and data leakage risks.

• Detecting AI usage does not automatically establish that a data breach has occurred.

• Enterprise AI accounts should be supported by clear governance, access controls, and data handling policies.

• DLP and appropriate AI security controls can help detect or prevent unauthorized transmission of sensitive information.

• Employee awareness and convenient approved tools are essential to sustainable AI security.