How to stop staff using public AI chatbots and use secure business AI instead
This article explains the risks associated with employees using public generative AI tools to process sensitive company and client information. It outlines how Australian small businesses can prevent accidental data leaks by implementing secure, enterprise-grade AI environments such as Microsoft Copilot with commercial data protection.
Understanding the risks of public AI chatbots
Many employees turn to free, public AI chatbots to draft emails, summarise documents, or write code. While these tools are convenient, they pose significant risks to small businesses. When staff upload client contracts, financial spreadsheets, or internal strategy documents into a public chatbot, that information may be stored on external servers and potentially used to train future AI models.
For Australian businesses, this can lead to breaches of the Privacy Act 1988 and the Australian Privacy Principles (APPs). Once sensitive data leaves your controlled environment, you lose visibility over who can access it and how it is used, creating legal and reputational exposure.
Step 1: Establish a clear AI usage policy
Before implementing technical controls, your business needs a formal policy that defines acceptable AI use. This document should be shared with all staff and included in your employee handbook.
- Clearly list which AI tools are prohibited for company use.
- Define what types of data must never be entered into a public chatbot (e.g., client personal information, financial records, intellectual property).
- Specify the approved business AI alternative and how to access it.
- Outline the process for requesting new AI tools through IT management.
Step 2: Deploy Microsoft Copilot with commercial data protection
Microsoft Copilot with commercial data protection provides a secure alternative to public chatbots. When staff sign in with their work account, prompts and responses are not saved, Microsoft has no eyes-on access to the data, and the information is not used to train the underlying foundation models.
To deploy this across your small business:
- Log in to the Microsoft 365 admin centre at
admin.microsoft.com. - Navigate to Billing > Your products and ensure you hold an eligible Microsoft 365 business licence that includes commercial data protection.
- Navigate to Users > Active users.
- Select the user you wish to licence, click the Licences and apps tab, and assign the appropriate licence.
- Click Save changes.
- Direct employees to access Copilot via
m365.cloud.microsoftor the Microsoft Edge sidebar using their work credentials. - Verify that staff are signing in with their business email address, not a personal Microsoft account.
Step 3: Configure Microsoft Purview sensitivity labels
Sensitivity labels allow you to classify and protect data across your Microsoft 365 environment. Follow these steps to configure labels in the Microsoft Purview compliance portal:
- Navigate to the Microsoft Purview compliance portal at
compliance.microsoft.com. - In the left-hand navigation pane, select Solutions > Information protection.
- Click Create a label to start the New sensitivity label wizard.
- Enter a name for the label (e.g., "Confidential - Client Data") and provide a description for staff.
- Choose the scope for the label. Select Items to apply it to emails and documents.
- Configure protection settings. You can choose to encrypt content or add visual markings such as headers and footers.
- Select Auto-labeling for files and emails if you want the system to apply the label based on detected sensitive information types (e.g., Tax File Numbers or credit card numbers).
- Review your settings and click Create label.
- Publish the label by creating a label policy and assigning it to specific users or groups.
Step 4: Manage internal permissions in SharePoint and OneDrive
Secure business AI is only effective if your underlying data permissions are correct. Copilot and similar tools can only access content that the user is already permitted to see. If your SharePoint sites and OneDrive libraries have overly broad permissions, the AI may surface sensitive information to the wrong employees.
Small businesses should:
- Audit SharePoint site permissions and remove "Everyone" or "All Staff" access to folders containing HR, payroll, or client data.
- Apply the sensitivity labels created in Microsoft Purview to classify documents and apply automatic protection policies.
- Restrict external sharing on sites that contain confidential client information.
- Regularly review guest access and sharing links to ensure former employees or external contractors no longer have access.
Step 5: Block access to unauthorised AI websites
To reduce temptation and accidental misuse, your IT provider can configure web filtering to block known public AI chatbot domains on business networks and managed devices. This does not replace staff training, but it adds a technical guardrail that prevents data from being pasted into unapproved tools.
Work with your MSP to implement DNS filtering using one of the following methods:
- DNS Filtering Service: Configure your business DNS forwarder to use a security-focused DNS provider. Create a custom block list containing domains associated with public generative AI chatbots.
- Firewall Rules: Configure outbound HTTP/HTTPS rules on your perimeter firewall to deny traffic to known AI chatbot IP ranges and domain names.
- Endpoint Security: Deploy a web filtering agent on managed devices to block access to unauthorised AI sites regardless of the network the device is connected to.
Ensure that access to your approved business AI platform (m365.cloud.microsoft) is explicitly allowed in your allow-list to prevent accidental blocking.
Step 6: Train staff on secure AI practices
Technology alone cannot prevent every data leak. Staff must understand why the policy exists and how to use the approved tool correctly. Conduct short training sessions covering:
- The difference between public chatbots and enterprise AI with commercial data protection.
- How to confirm they are signed into the business version of Copilot.
- What information remains safe to process in the approved environment.
- How to report accidental data entry into a public AI tool.
Maintaining compliance with Australian privacy obligations
Using secure business AI helps your organisation meet its obligations under the Privacy Act 1988. By keeping data within your Microsoft 365 tenant and ensuring it is not used to train public models, you maintain greater control over personal information and reduce the risk of notifiable data breaches. Document your AI governance approach as part of your broader privacy and security framework.
Troubleshooting common issues
Staff cannot access Microsoft Copilot
- Issue: User receives an access denied message or is redirected to a personal account.
- Resolution: Verify that the user has been assigned an eligible licence in the Microsoft 365 admin centre under Users > Active users > Licences and apps. Ensure the user is signing in with their business email address and not a personal Microsoft account (e.g., @outlook.com or @hotmail.com).
Licence assignment errors
- Issue: Error message when attempting to assign a licence to a user.
- Resolution: Check that your tenant has available seats for the specific licence type. Navigate to Billing > Licences in the Microsoft 365 admin centre to purchase additional licences if necessary. Ensure the user's location is correctly set in their profile, as licence availability varies by region.
Sensitivity labels not appearing in Office apps
- Issue: Labels created in Microsoft Purview are not visible in Word, Excel, or PowerPoint.
- Resolution: Sensitivity label propagation can take up to 24 hours. Ensure the user has restarted their Office applications. Verify that the label policy has been published to the specific user or group in the Microsoft Purview compliance portal under Information protection > Label policies.
Blocked access to approved business AI
- Issue: Staff cannot reach
m365.cloud.microsoft after DNS filtering is implemented.
- Resolution: Check your DNS filtering block lists and firewall rules to ensure the domain
m365.cloud.microsoft and its associated subdomains are added to the allow-list. Verify that the endpoint security policy on the device is not overriding the network allow-list.
m365.cloud.microsoft after DNS filtering is implemented.m365.cloud.microsoft and its associated subdomains are added to the allow-list. Verify that the endpoint security policy on the device is not overriding the network allow-list.