Can employees use ChatGPT, Copilot or Claude with internal systems safely?
Short answer: yes, when the connection between the AI app and your systems carries each employee's own identity, is limited by rules you set, and is recorded. Desktop and IDE apps that let each person add their own tool server (Claude Desktop, Cursor, VS Code, Open WebUI) do this well. Team-wide connectors configured once by an admin usually hold a single key, which makes everyone's AI act as one account, so check that before you connect HR or finance.
The safe pattern
- One door: the AI reaches your systems through one gateway, never through a pile of direct connectors.
- Each person's own token: the AI app sends a personal token, so the gateway knows exactly who is asking.
- Their own access at the far end: calls run with that person's own login to the HR system, the CRM or the file share.
- Rules and a record: deny by default, approvals for changes, and every call in your SIEM.
Which AI apps fit the pattern
- Claude Desktop, Cursor and VS Code: each person adds the gateway as an MCP server with their own token. Good fit.
- Open WebUI with a local model: each person's token, or the app vouches for the signed-in person as a trusted client. Good fit, and nothing leaves the building.
- ChatGPT custom connectors and Microsoft Copilot Studio: they can add remote MCP servers, but the connection is set up once and runs from the vendor's cloud. A single API key there means one identity for everyone, and the server must be reachable from the internet. Use them for systems where one shared identity is acceptable, or only where each person signs in individually. Check your plan's current options, since they change often.
Connect an AI app to Acutis Gate
Acutis Gate gives every employee one endpoint and a personal token.
- In the Gate console, open Connect an AI, name the token after the app (for example "Claude Desktop") and click Create token. The page shows a ready-made config for each app with a Copy button.
- Paste it into the app: Claude Desktop's config file, Cursor's
mcp.json, or Open WebUI's External Tools. - Restart the app and ask something only your role should be able to answer. Then check Audit trail: the row names you.
The Cursor config, as Gate generates it:
{
"mcpServers": {
"acutis-gate": {
"url": "https://gate.corp.example/api/v1/gate/mcp",
"headers": { "Authorization": "Bearer <your personal token>" }
}
}
}
Claude Desktop reaches remote servers through the mcp-remote helper, and the console's copy button writes that version for you. A token can be revoked at any time, and a revoked token that is still in use shows up in the trail as a failed sign-in.
Any AI app, each person's own access
Gate gives every employee one endpoint and a personal token, and runs each call with their own permissions.
Start a 14-day trial Tour the live GateFrequently asked questions
Is it safe to connect ChatGPT or Copilot to internal company systems?
It can be, when each connection carries the employee's own identity, rules limit what it can do, and every call is recorded. Connectors configured once with a single API key make everyone act as one account, so avoid them for sensitive systems.
Which AI apps support a personal token per employee?
Apps where each person adds their own MCP server, such as Claude Desktop, Cursor, VS Code and Open WebUI, work well with a personal token per employee.
Do ChatGPT connectors work with servers inside my network?
Cloud-run connectors call your server from the vendor's cloud, so the server must be reachable from the internet. Desktop apps and Open WebUI can reach a gateway inside your network.
How do I see which AI app made a call?
Gate records the client with each call, along with the person, the tool, the arguments and the result.
Acutis