Short answer
Combine four layers: give staff an approved business AI tool so they do not need consumer ones, publish a clear policy on what data may go into AI, enforce it with Microsoft Purview DLP on endpoints and browsers plus Defender for Cloud Apps to block or monitor unapproved AI sites, and train people with real examples. Blocking alone pushes use to personal phones.
Key facts
- Most leakage happens in consumer AI accounts used because no approved tool exists.
- Business AI plans such as Microsoft 365 Copilot and Claude business plans come with commercial data terms and admin control.
- Endpoint DLP in Microsoft Purview can warn on or block pasting and uploading sensitive content to specified websites.
- Defender for Cloud Apps can discover AI apps in use and mark unapproved ones as unsanctioned.
- A policy people understand prevents more leaks than a block list they work around.
Why blocking alone fails
Staff use AI tools because they help. If the company offers nothing approved, people use free consumer accounts on work laptops, or copy data to their phones. Blocking every AI website on the network moves the problem out of view instead of solving it.
The approach that works is to make the safe path the easy path: provide an approved tool, explain the rules, enforce them on the data that matters most, and keep watching what is in use.
Layer 1: Provide an approved AI tool
Pick one or two business AI tools, such as Microsoft 365 Copilot Chat, Microsoft 365 Copilot or a Claude business plan, and make them available to the staff who need them. Business plans come with commercial terms on data use, sign-in through Entra ID, and admin controls. Microsoft 365 Copilot Chat with enterprise data protection is available to many Microsoft 365 users when signed in with a work account.
Tell people clearly which tools are approved and how to get access. The fewer reasons they have to use a personal account, the fewer leaks you need to catch.
Layer 2: A clear AI data policy
Write rules in plain language, grouped by data type rather than by tool. Include it in the AI acceptable use policy and onboarding.
| Data type | Approved business AI tool | Unapproved or personal AI tool |
|---|---|---|
| Public information | Allowed | Allowed |
| Internal documents | Allowed | Not allowed |
| Customer personal data | Allowed only for approved use cases | Not allowed |
| Health, payment or government ID data | Only with compliance approval | Not allowed |
| Passwords, keys, credentials | Not allowed | Not allowed |
| Source code and trade secrets | Per policy and tool approval | Not allowed |
Layer 3: Technical enforcement
Microsoft Purview gives several controls. Endpoint DLP on onboarded Windows and macOS devices can audit, warn or block when users paste or upload content matching sensitive information types or sensitivity labels to specified domains, such as consumer AI sites. Sensitivity labels let you target the content you care most about. Purview also offers AI-focused views, such as Data Security Posture Management for AI, to see interactions with AI apps where licensed.
Defender for Cloud Apps discovers which cloud and AI apps are in use from endpoint or network signals, and can mark unapproved ones as unsanctioned so Defender for Endpoint blocks them. Conditional Access and Intune app protection keep company data in managed apps on mobile devices. Licensing for these features varies by plan, so map controls to what you own.
- Onboard devices to Purview endpoint DLP and start in audit mode.
- Create a list of AI sites to warn or block for sensitive content.
- Use Defender for Cloud Apps discovery to find shadow AI.
- Mark clearly unapproved AI apps as unsanctioned.
- Move from audit to warn to block as false positives are tuned.
Layer 4: Training and feedback
Short training with real examples works best: a customer list pasted into a chatbot, a contract uploaded for summary, a script with a hardcoded password. Show the approved alternative for each. DLP policy tips that explain why an action was blocked are training in the moment.
Review DLP alerts and Cloud Apps discovery monthly. Repeated warnings for the same team usually mean they need an approved tool for a real task, not more blocking.
How Accred Consulting approaches it
Accred Consulting configures layered AI data controls in Microsoft 365; the client and its counsel own the policy and compliance decisions. Discovery reviews which AI tools are in use, sensitive data types, licensing and device management, and the proposal sets scope, fixed price and timeline. The US-based specialists assigned to your engagement deploy approved AI tools such as Microsoft 365 Copilot or Claude business plans, configure sensitivity labels, Purview endpoint DLP and Defender for Cloud Apps discovery, and help draft the AI data policy. Handover includes documentation and knowledge transfer, with optional support retainers.
Track record: 475+ client projects, 2,000+ migration workstreams and 210,000+ users migrated by a US-based team. Company facts · Case studies
Related questions
Want this scoped for your environment?
Book a free consultation. After discovery you get a written plan with scope, a fixed price and a timeline.
Book a free consultation