At least a few people in your organization started using Claude in the last few months. Some are running it through a personal account. A few are pasting content from your M365 environment directly into the chat to get a faster draft. And in most cases, IT has no visibility into any of it.
IBM’s 2025 Cost of a Data Breach Report found that only 37% of organizations have any AI governance policy at all. The Verizon 2026 DBIR documented a tripling of corporate AI usage in a single year, with two-thirds of that activity happening through accounts the organization doesn’t control. Claude specifically has gained fast traction in nonprofit, association, and document-heavy professional environments.
Claude Isn’t Just a Chatbot Anymore
Claude runs across three distinct surfaces and they carry very different security implications:
- The web interface (claude.ai): The most common entry point. Staff type or paste content directly into the browser. Client information, internal documents, financial data, and donor records can all leave the organization through what looks like normal browsing activity.
- The Claude desktop app: Bypasses browser-based controls entirely. Files dragged in or clipboard content pasted into the desktop app falls outside the reach of most standard DLP tools.
- MCP integrations: Model Context Protocol lets Claude connect directly to Slack, Google Drive, M365 SharePoint, and internal databases. When those connections exist, Claude isn’t just answering questions. It’s operating as a privileged user across your systems, pulling data autonomously. MCP adoption grew over 400% in 2025, largely without security oversight.
The Setup Decisions That Determine Whether Claude Is Safe
Anthropic’s own framework draws a clear line: they’re responsible for securing the model. Your organization is responsible for everything else, meaning how it’s accessed, what data flows through it, and who can use it.
A few decisions that carry the most weight:
- Identity and access: Tying Claude to your SSO/SAML provider means access is automatically revoked when staff leave. Without it, Claude accounts can persist well after offboarding.
- API key management: Shared team API keys are a liability. Keys should be scoped per user or function, stored in a secrets manager, and rotated on schedule. Keys living in environment files or committed code repositories are a known attack vector.
- MCP inventory: Every MCP integration is a new attack surface. A compromised or malicious MCP server can inject instructions directly into Claude’s context, a technique called prompt injection. Without an inventory of approved connections, there’s no baseline to audit against.
- Audit logging: AI interactions fall outside traditional monitoring workflows by default. That gap needs to be deliberately closed with logging flowing into your SIEM or security monitoring environment.
What a Properly Governed Deployment Actually Looks Like
The goal isn’t to slow adoption. It’s to make adoption sustainable. Organizations that get this right treat AI governance as infrastructure, not a policy document that lives in a folder.
In practice, that means centralized access tied to corporate identity with automatic offboarding, a data classification policy that tells staff what can and can’t go into AI interfaces, an inventory of every integration and connected data source, and audit logging that creates visibility rather than assumptions.
IBM’s breach data puts the premium for shadow AI incidents at $670,000 above a standard breach. That’s the gap between having governance in place and not having it.
How Orion Networks Helps Organizations Set This Up Properly
Orion Networks works with nonprofits and associations across the DC area to build IT environments that can absorb new technology without creating new risks. AI governance is now part of that work.
If your organization is using Claude, thinking about adopting it formally, or concerned about what your staff is already doing with personal accounts, we can help you understand the exposure and build a deployment that fits your security posture.
It all starts with a conversation. Reach out to the Orion Networks team at (202) 505-6157 to talk through where you are.
