What employees will see

Once InPolicy is deployed to an AI assistant, employees don’t have to do anything differently. But they will notice a few things.

What they’ll notice

Occasional policy callouts in AI responses. When a company policy is relevant to what they asked, the AI may explicitly mention it:

“I can help you draft that customer email. Per our external communications policy, I’ll avoid naming specific enterprise accounts. Let me know if you want me to handle that differently.”

This is normal and expected. It’s the AI being transparent about what’s guiding its response.
Some requests getting redirected or partially declined. If something they ask would clearly violate a high-severity policy, the AI will decline that specific part while offering alternatives:

“I can outline the Q4 strategy in general terms, but the specific pipeline numbers should come from your manager. They’re covered by our confidentiality policy. Here’s a framework you could use for the rest…”
A very small latency bump (typically under a second) on the first message of a conversation, and imperceptible on subsequent turns.

What they won’t see

Any new UI. The AI assistant looks identical. No InPolicy banners, popups, or indicators.
Their conversations being stored anywhere. InPolicy does not store the text of what they or the AI say. See Privacy.
Per-user tracking. InPolicy does not know who is using the AI; only that an InPolicy-enabled conversation is happening somewhere in your tenant.

What to tell them

A one-paragraph email or Slack message is usually enough:

Hi team, we’ve enabled InPolicy in Claude Desktop / Cursor / Claude Code. This means the AI is now aware of our company’s policies and will follow them on your behalf. You might notice the AI mentioning specific policies in its responses, or redirecting requests that would violate one. No change to how you use the tools. Questions → #it-help.

FAQ to include

Q: Is my conversation being logged?

A: No. InPolicy does not store the content of your messages. It only tracks which policies were active during the conversation and for how long. See the privacy details.

Q: Can I turn it off if it gets in the way?

A: Yes, using your AI assistant’s own settings, not anything in InPolicy. In Claude Desktop, go to Settings → Developer and toggle the inpolicy server off. In Cursor, Settings → MCP Servers. In Claude Code, claude mcp disable inpolicy. Don’t leave it off; the policies exist for a reason. Short-term overrides for legitimate work are fine. Your IT admin can see which servers you have disabled via your assistant’s telemetry, not via InPolicy.

Q: Does it work with ChatGPT, Gemini, Copilot, etc.?

A: Currently InPolicy integrates with MCP-compatible assistants (Claude Desktop, Claude Code, Cursor, Windsurf). OpenAI-family products don’t support MCP yet; IT can build custom integrations using the REST API if needed.

Q: I noticed the AI cited a policy that doesn’t seem to apply. What do I do?

A: Take a screenshot and send it to IT or your policy owner. Occasionally the semantic search will surface a tangentially-related policy. Those reports help tune the retrieval over time.