Stop Your Staff from Pasting Client Data into Public AI Tools: A Guide to Shadow AI and Data Leak Prevention
Summary
This guide helps you troubleshoot and resolve: Stop Your Staff from Pasting Client Data into Public AI Tools: A Guide to Shadow AI and Data Leak Prevention. Follow the steps below to fix the issue.
Common Causes
Shadow AI occurs when employees use consumer-grade artificial intelligence tools—such as public chatbots and image generators—without organisational oversight. When staff paste client records, proprietary code, or financial spreadsheets into these platforms, that data often leaves your secure environment and may be used to train public models. For Australian businesses, this creates significant risks under the Privacy Act 1988 and the Australian Privacy Principles (APPs), potentially leading to notifiable data breaches under the Notifiable Data Breaches (NDB) scheme and loss of client trust.
This article is categorised under Security & Access due to its focus on data leak prevention, Shadow AI mitigation, and Privacy Act compliance. If your MSP prefers Cloud Migration & Modern Workplace, please note that the content is primarily security and governance focused.
Quick Fix Steps
- Audit current AI usage by reviewing web filtering logs, browser history, and Microsoft Intune endpoint reports to identify visits to consumer AI platforms.
- Block unauthorized AI domains at the network level using your firewall or DNS filtering solution, and at the endpoint level via Microsoft Intune configuration profiles.
- Deploy Microsoft Copilot with commercial data protection by assigning appropriate Microsoft 365 licenses and configuring settings in the Microsoft 365 admin center.
- Train employees on prohibited data types and require sign-in with organisational credentials at copilot.microsoft.com.
- Update your acceptable use policy to explicitly cover generative AI and BYOD scenarios.
Detailed Instructions
1. Audit Current AI Usage
Review your web filtering logs, browser history, and Microsoft Intune endpoint management reports to identify visits to consumer AI platforms. Common indicators include frequent access to public chatbot domains and large file uploads to unknown external sites. Speak with team leads to understand which departments are experimenting with AI and for what purposes. Document all unauthorized tools discovered during this audit.
2. Restrict Access to Public AI Tools
Use your firewall, DNS filtering service, or endpoint management console to block known consumer AI domains. For Microsoft Intune, follow these steps:
- Sign in to the Microsoft Intune admin center at endpoint.microsoft.com.
- Navigate to Devices > Configuration profiles > Create profile.
- Select Windows 10 and later as the platform and Settings catalog as the profile type.
- Search for and add the Web browser settings category.
- Configure URL blocklist to include known consumer AI domains and URL allowlist for approved enterprise AI endpoints.
- Assign the profile to the appropriate user or device groups and select Save and Deploy.
For firewall or DNS filtering, add consumer AI domains to your block list and ensure the policy applies to both wired and wireless networks. Update the block list regularly as new AI tools emerge.
3. Deploy Enterprise-Grade AI with Commercial Data Protection
Provide staff with a secure, managed AI environment. Microsoft Copilot with commercial data protection ensures that user and organisational data is not used to train the underlying public foundation models, provided users sign in with their organisational Microsoft 365 account.
- Sign in to the Microsoft 365 admin center at admin.microsoft.com.
- Navigate to Billing > Licenses and assign Microsoft 365 E3, E5, Business Premium, or Copilot licenses to your users.
- Navigate to Settings > Org settings > Services > Copilot.
- Enable Commercial data protection for your organisation and select Save.
- Instruct staff to sign in at copilot.microsoft.com using their work email address, not a personal Microsoft account.
- Verify that the protected badge or enterprise sign-in indicator appears before any data is entered.
4. BYOD and Personal Device Risks
Employees may attempt to bypass network restrictions by using personal smartphones, tablets, or laptops to access public AI tools, or by connecting company devices to mobile hotspots. This creates unmanaged data exfiltration paths outside your visibility.
- In Microsoft Intune, navigate to Devices > Compliance policies > Create policy.
- Select the platform (e.g., Android, iOS/iPadOS, or Windows) and configure settings to require device encryption and a minimum OS version.
- Enable Conditional Access in the Microsoft Entra admin center to block access to organisational resources from non-compliant or unenrolled personal devices.
- For unmanaged devices, implement App Protection Policies in Intune under Apps > App protection policies to prevent copy-paste of organisational data into consumer AI applications.
- Update your acceptable use policy to prohibit the use of personal devices and mobile hotspots for processing client or confidential business data.
5. Establish Data Handling Rules
Create clear rules about what information may be processed by AI tools. As a baseline, prohibit pasting the following into any AI platform unless it is your approved enterprise environment:
- Client names, contact details, and personal information
- Customer databases, invoices, or financial records
- Proprietary source code or intellectual property
- Employee records and payroll information
- Confidential contracts or legal documents
6. Update Your Acceptable Use Policy
Revise your organisation's acceptable use and IT security policies to include generative AI. The policy should define approved tools, prohibited data types, BYOD restrictions, and the process for requesting new AI software. Require all staff to acknowledge the updated policy, and include AI awareness in your onboarding and annual security training.
Troubleshooting
Staff Continue Accessing Blocked AI Sites
If users are bypassing network blocks, check for personal mobile devices on your Wi-Fi, VPN tunnels, or proxy services. Ensure your filtering solution applies to both wired and wireless networks, and consider deploying DNS-level protection that covers roaming devices. Review Intune compliance policies to confirm personal devices are not accessing corporate resources.
Employees Are Unsure Which AI Tool Is Approved
Publish a simple internal page listing the approved enterprise AI platform and how to access it. Include screenshots showing the correct sign-in screen and the commercial data protection indicator so staff can verify they are in the secure environment before starting work.
Staff Use Personal Devices to Access AI Tools
If staff use personal phones or laptops to access public AI platforms with client data, treat this as a shadow IT incident. Identify the device and user, remind them of the acceptable use policy, and enforce Conditional Access rules to block organisational data access from unenrolled devices. If organisational data was entered, proceed to the breach response steps below.
Enterprise AI Tool Is Not Licensed or Available
If your organisation does not have Microsoft 365 licenses that include Copilot with commercial data protection, prohibit all use of generative AI tools for client or confidential business data until an approved solution is procured. Contact your MSP to discuss enterprise AI licensing options or alternative secure platforms that meet Australian privacy requirements.
Data May Have Already Been Pasted into a Public AI Tool
If you suspect client data has been entered into a consumer AI platform, treat it as a potential notifiable data breach under the Privacy Act 1988.
- Contain the incident by revoking the user's access to the public AI platform and changing any compromised credentials.
- Document what information was shared, when, by whom, and whether the data includes personal information as defined by the APPs.
- Assess whether the incident is likely to result in serious harm to affected individuals. Consider the sensitivity of the data, whether it was encrypted or anonymised, and whether the AI provider retains or uses the data for model training.
- If the breach meets the notifiable threshold, notify the Office of the Australian Information Commissioner (OAIC) and affected individuals as soon as practicable. Refer to the OAIC Notifiable Data Breaches scheme guidance at oaic.gov.au.
- Seek legal advice if client personal information may have been exposed, and review your internal controls to prevent recurrence.
For assistance with implementing these controls or responding to a suspected data breach, Open a Ticket.
Still Having Issues?
💻 Open a Ticket