AI Governance Controls

How to enforce approved AI tool usage and prevent employees from using unsanctioned AI services.

Why shadow AI is a risk

When employees use personal AI accounts, company data leaves your enterprise boundary — no audit trail, no data residency guarantees, no DLP controls, and no way to revoke access when someone leaves. Claude Enterprise keeps data within your agreed data processing terms; personal accounts do not.

Four enforcement layers — use all for strongest coverage
Policy
Mandatory, immediate
Network
Blocks on corporate Wi-Fi
Endpoint
Blocks on managed devices
CASB
Strongest, full visibility
Layer 1 — Acceptable Use Policy
Mandatory baseline before any technical controls
Name Claude Enterprise as the only approved AI assistant
Explicitly list prohibited tools (ChatGPT, Gemini, Copilot personal, etc.)
State consequences: disciplinary action for data exfiltration via personal AI
Require employees to acknowledge and sign annually
Policy alone is weak enforcement — determined users bypass it. Combine with at least one technical layer. Policy is necessary for disciplinary action when technical controls are circumvented.
Layer 2 — Network / DNS filtering
Blocks access on corporate network and VPN
Domains to block
chat.openai.comchatgpt.comgemini.google.combard.google.comcopilot.microsoft.combing.com/chatperplexity.aicharacter.aimeta.aipoe.comclaude.ai (personal)you.com
Tools: Cisco Umbrella, Cloudflare Gateway, Zscaler DNS, pfSense blocklist
Allow Enterprise subdomain while blocking personal claude.ai (tenant-level split)
Does NOT block mobile data — employees can use 4G/5G to bypass
Does NOT block personal VPNs unless those are also blocked
Layer 3 — Endpoint / MDM controls
Applies to managed devices even off corporate network
Windows (Intune / GPO)
1Force corporate account in Chrome/Edge — BrowserSignin policy
2Disable personal browser profiles
3Deploy URL blocklist via Chrome Enterprise or Edge policy
4Always-on VPN forces all traffic through corporate DNS
macOS (Jamf / Mosyle)
1Manage Safari via content filter profile
2Push Chrome enterprise config via Jamf policy
3Block app installs from unmanaged sources
4Always-on VPN (e.g. Jamf Connect) for DNS enforcement
Endpoint controls only cover managed devices. BYOD and unmanaged devices need network or CASB controls.
Layer 4 — CASB (strongest)
Cloud Access Security Broker — full visibility and real-time enforcement

A CASB sits between users and the internet (via agent or proxy), inspects HTTPS traffic, and can block specific SaaS applications even when accessed on mobile data or personal devices — while leaving approved apps accessible.

Blocks shadow IT AI in real time with user notification
Logs all AI tool usage attempts — audit trail for compliance
Can allow Claude Enterprise tenant while blocking personal claude.ai logins
Covers BYOD and mobile when agent is installed
Zscaler Internet Access
Netskope
Microsoft Defender for Cloud Apps
Claude Enterprise: enforce SSO

In your Claude Enterprise Admin Console, enforce SSO so employees must authenticate via your corporate IdP (Okta, Azure AD, Google Workspace). This means a personal claude.ai account cannot be used with a corporate email — the IdP controls access. Combined with network blocking of the personal claude.ai endpoint, only your Enterprise tenant remains reachable.

1Admin Console → Security → SSO
2Configure your IdP (SAML or OIDC)
3Set SSO enforcement to Required (not optional)
4Block personal claude.ai at network layer — Enterprise subdomain stays open
Recommended rollout order
Week 1Publish AUP. Announce Claude Enterprise as the only approved tool. Communicate why.
Week 2Enforce SSO in Claude Enterprise Admin Console.
Week 3Deploy DNS/network blocklist for known AI domains.
Month 2Roll out endpoint browser policies via MDM for managed devices.
OngoingReview CASB reports (or network logs) monthly for new AI domains employees attempt to reach. Add to blocklist.
Spending ControlsAnthropic: SSO setup guide