Five Steps to Manage Shadow AI Without Slowing Employees

Summary
The sponsored guide recommends inventorying AI tools and their OAuth permissions, setting approved-tool and data-use policies, speeding reviews, monitoring browser activity, and coaching employees to reduce shadow AI exposure.
Key points
- Unapproved AI tools may gain access to corporate drives, email, or documents through OAuth permissions or browser sessions.
- Inventory OAuth-connected apps, browser extensions, and AI features added to already-approved products; employee surveys can reveal tools automated discovery misses.
- AI policies should identify approved tools, specify data that must not be entered, verify training opt-out settings, and explain how employees can request new tools.
- A structured review process can assess data access, vendor security practices, training opt-outs, certifications, and available approved alternatives.
- The article recommends browser-native monitoring to give security teams visibility into AI use and alert employees to potential risks.
- Just-in-time coaching and training that explains the reasons behind policies can direct employees to approved alternatives and help them make safer choices.
Article Details
- Defense Focus
- Improve visibility and governance of employee use of unapproved AI tools, especially tools that access corporate data through OAuth or browser sessions.
- Detection Methods
- Quarterly audits of connected third-party OAuth apps, sorted by permission scope.
- Scan employee devices for active browser extensions using browser management or a lightweight agent.
- Use employee surveys to identify AI tools that automated discovery may miss.
- Monitor AI activity with a browser-native approach and correlate it with phishing simulation results and training completion data.
- Data Sources
- Third-party OAuth app connections and their permission scopes
- Active browser extensions on employee devices
- Employee survey responses about AI tool use
- AI activity monitoring signals
- Phishing simulation results and training completion data
- Platforms
- Browser-native monitoring
- Browser management solutions
- Endpoint management tools
- Defensive Actions
- Maintain an inventory of AI tools in use, their users, and the data they can access.
- Publish a current list of approved tools and provide a clear, timely process for requesting new ones.
- Define data classification rules for information that must not be entered into AI tools.
- Verify that approved tools handling sensitive data are configured to opt out of using company inputs for model training.
- Evaluate new tools based on data access scope, vendor security practices, training opt-out status, compliance certifications, and existing approved alternatives.
- Use just-in-time coaching to explain risks and direct employees attempting to use unsanctioned tools to approved alternatives.
Vendors
Products
Google GeminiMicrosoft Copilot, Google Gemini, and Salesforce Einstein are examples of AI capabilities that may have been introduced after the original vendor review, often without a separate security evaluation.Google WorkspaceMost AI tools request access to Google Workspace or Microsoft 365 through OAuth, which grants them read or write permissions to corporate data.Microsoft 365 Apps for EnterpriseMost AI tools request access to Google Workspace or Microsoft 365 through OAuth, which grants them read or write permissions to corporate data.Microsoft CopilotMicrosoft Copilot, Google Gemini, and Salesforce Einstein are examples of AI capabilities that may have been introduced after the original vendor review, often without a separate security evaluation.Salesforce EinsteinMicrosoft Copilot, Google Gemini, and Salesforce Einstein are examples of AI capabilities that may have been introduced after the original vendor review, often without a separate security evaluation.
