Full Report
When an employee installs an AI writing assistant, connects a coding copilot to their IDE, or starts summarizing meetings with a new browser tool, they are doing exactly what a productive employee should do: finding faster ways to work. Across most organizations today, employees are running three to five AI tools on any given day. Most were never reviewed by IT. A significant portion connects
Analysis Summary
# Best Practices: Managing Shadow AI Tools
## Overview
These practices address the "Shadow AI Gap"—the security risk created when employees use unreviewed AI writing assistants, coding copilots, and browser tools that connect to corporate data via OAuth or browser sessions. These guidelines aim to provide visibility and control without hindering employee productivity.
## Key Recommendations
### Immediate Actions
1. **Conduct a Multi-Channel Inventory:** Identify AI tools currently in use by auditing OAuth connections in Google Workspace/Microsoft 365, scanning browser extensions, and reviewing AI features bundled in existing software (e.g., Salesforce, Zoom).
2. **Deploy an Employee Survey:** Ask employees which AI tools they use to save time. Frame the survey as an effort to "support safe productivity" rather than a disciplinary audit.
3. **Implement Data Training Opt-Outs:** For all known AI tools, manually configure settings to ensure corporate data is not used to train the vendor’s public models.
### Short-term Improvements (1-3 months)
1. **Draft a Functional AI Policy:** Create a "live" document that lists approved tools, clear data classification rules (what *cannot* be entered into AI), and a plain-language explanation of risks.
2. **Establish a "Fast Lane" Review Process:** Create a streamlined procurement/review path specifically for AI tools to prevent employees from bypassing IT due to slow approval times.
3. **Define Data Guardrails:** Categorize sensitive data (PII, source code, financial records) and establish strict "No-Go" zones for non-enterprise AI versions.
### Long-term Strategy (3+ months)
1. **Continuous Monitoring Integration:** Move from quarterly audits to automated monitoring of OAuth permissions and browser-based data exfiltration to AI domains.
2. **Continuous Education Program:** Transition from a static policy to regular security awareness training focused on the evolving nature of Generative AI risks.
3. **Consolidate AI Tooling:** Gradually migrate users from disparate "Shadow AI" tools to a standardized suite of enterprise-grade AI tools with centrally managed security controls.
## Implementation Guidance
### For Small Organizations
- **Focus on inventory:** Use manual audits of Google/Microsoft admin consoles to revoke high-risk OAuth permissions.
- **Leaning on Policy:** Focus on a strong "Acceptable Use Policy" and clear verbal communication.
### For Medium Organizations
- **Automated Discovery:** Use browser management tools or lightweight endpoint agents to track AI browser extensions.
- **Formal Intake:** Use a simple ticketing system or form to centralize and speed up AI tool requests.
### For Large Enterprises
- **Governance Frameworks:** Formalize a "Fast Lane" committee involving Legal, Privacy, and Security to review tools in days, not months.
- **CASB/SASE Integration:** Use Cloud Access Security Brokers (CASB) to block unauthorized AI domains while allowing approved ones.
## Configuration Examples
- **OAuth Scoping:** Restrict "Read/Write" access for 3rd-party AI apps to specific, non-sensitive directories rather than full drive/inbox access.
- **Enterprise Toggle:** In tools like ChatGPT Enterprise or Midjourney, ensure "Training on Business Data" is toggled **OFF** in the admin console.
- **Browser Management:** Use Chrome Enterprise or Edge management policies to block specific high-risk AI extensions while whitelisting approved assistants.
## Compliance Alignment
- **NIST AI Risk Management Framework (AI RMF):** Aligning with governance and map functions.
- **ISO/IEC 42001:** Establishing an Artificial Intelligence Management System (AIMS).
- **CIS Controls:** Specifically Control 2 (Inventory and Control of Software Assets) and Control 5 (Account Management).
## Common Pitfalls to Avoid
- **The "Wall of No":** Outright banning AI tools generally fails, as employees will simply find harder-to-detect ways to use them.
- **Ignoring Bundled AI:** Forgetting that "Approved" vendors (like Microsoft or Adobe) frequently add new AI features that may not fall under the original security assessment.
- **Default Settings:** Assuming a "Pro" or "Paid" account automatically protects your data from being used as training fodder; this often requires manual configuration.
## Resources
- **NIST AI RMF:** [https://www.nist.gov/itl/ai-risk-management-framework]
- **Gartner AI Governance Research:** [https://www.gartner.com/en/information-technology/topics/generative-ai]
- **OWASP Top 10 for LLMs:** [https://owasp.org/www-project-top-10-for-large-language-model-applications/]