Your privacy: who can see your conversations

Let’s answer the question you actually came here with: when I type something into Chat, who reads it?

The short version: your conversations belong to your organization, they stay inside your organization, and they are not added to any public AI’s training data. Within your organization, a conversation is yours by default — other people don’t browse it. A small set of trusted controls exist so that a regulated business can use AI responsibly, and this page explains exactly what they are, when they apply, and what they can and can’t see.

The one-sentence answer

Your conversations are private to you and your organization. They are never shared with public AI services, never used to train any public model, and never visible to other companies. Inside your organization, only people your administrators have given an oversight role — and only for legitimate compliance reasons — can review conversation activity, and every such review is itself recorded.

The rest of this page unpacks that sentence so you can trust it rather than just take our word for it.

What “private to your organization” really means

Chat is a business product, not a consumer app. When your company adopts Chat, it gets its own private workspace. Think of it as your company’s own locked building:

Your conversations, your uploaded documents, your notes, and the things Chat remembers about your work all live inside your company’s workspace.
Another company using Chat is in a completely separate building. They cannot see into yours, and you cannot see into theirs. There is no shared space where data from different companies mixes.
This separation is enforced by the system on every single request — not as an honor system, but as a rule the software applies to every read and every write. Independent isolation tests run against this continuously.

So the first and most important boundary is the one between organizations: what happens in your workspace stays in your workspace.

For a deeper look at how that wall between organizations is built and verified, see How your data stays isolated.

Who can see your conversations inside your own organization

This is the part most people really want to know. Here’s the honest, complete picture.

By default: just you

A conversation you start is yours. Your colleagues do not see your private conversations, and there is no “everyone’s chats” list that coworkers can scroll through. If you want to share a conversation, you choose to share it — it doesn’t happen on its own.

When you choose to collaborate

Some things are shared because you shared them:

If you share a conversation with a teammate or branch it into a shared project, the people you shared it with can see it. That’s the point of sharing.
If you work inside a shared project or workspace folder, the materials in that shared space are visible to its members.

In other words, the people who see your content are the people you invited into it.

People with an oversight role

Your organization’s administrators can grant a small number of people an oversight role for compliance and security purposes. The most relevant one is a read-only compliance/governance reviewer:

It is read-only. A reviewer in this role can look at activity records to confirm the AI is being used safely and within policy. They cannot edit your conversations, post as you, or change your settings.
It exists so that organizations in regulated industries — finance, healthcare, legal, and so on — can answer the question “is our team using AI responsibly?” without that turning into surveillance of individuals.
Granting this role is itself a controlled, logged action. Roles can only be assigned by administrators, and every assignment is recorded.

The principle here is separation of duties: the people who can review activity for compliance are deliberately not the same as the people who can change things, and neither group can quietly expand its own access. The rules that govern who can do what are built into the system and can’t be edited on the fly to grant someone extra reach.

The reassurance, plainly: the only people inside your organization who can review your conversation activity are people your own administrators have intentionally placed in an oversight role for compliance — and that review is itself recorded so it can be held accountable.

What “monitored” means here (and what it doesn’t)

You may have heard that AI usage at work is “monitored.” That word can sound ominous, so let’s be precise about what it actually means in Chat.

“Monitored” means there is an automated safety review and a compliance record — not a person reading over your shoulder.

Two distinct things are happening, and neither is a human silently watching you type:

An automated safety review. Messages pass through an automated AI-safety check as part of normal processing — the same way an email system scans for spam and malware without a person reading your email. This is software working automatically in the background, and it’s there to catch genuinely harmful or policy-violating content, not to judge your everyday work. If something is ever held back, you’ll see a clear notice explaining that your message was reviewed for safety — you can read more in Why it said no.
A compliance record. The system keeps records that an organization needs to demonstrate responsible AI use — for example, that the safety review ran. These records support audits and accountability. They are not a browsable feed of everyone’s chats, and access to them is itself a controlled, logged privilege.

So “monitored” is best read as “governed and accountable,” not “spied on.” The monitoring is mostly machinery doing its job quietly; the human element is limited, role-gated, and logged.

This is also why Chat is genuinely safe to use for real work — Safe at work covers that from the IT and trust angle.

Your data is not shared with public AI

This is a worry people bring from consumer chatbots, and it deserves a flat, direct answer.

Your conversations are not used to train any public AI model. What you type does not become training data for a public model that other people then benefit from. Your content works for you and your organization, full stop.
Your data does not leave your organization’s boundary to a public chatbot service. Chat is not a front-end for a consumer chatbot account where your prompts pile up in someone’s public service.
The AI does not browse the open internet with your data unless you ask it to. Web search is an opt-in, approval-gated capability — it doesn’t quietly send your conversation out to the open web. When information does come in from a search, it passes through the same safety review as everything else before it’s used.

If you’ve used a free consumer chatbot before, this is one of the biggest differences — How Chat is different from other chatbots draws out the rest of the contrast.

A quick mental model

Picture your message making a short, supervised trip and coming straight back:

You type a message
        │
        ▼
  Automated safety review  ──► (clear notice if anything is held back)
        │
        ▼
  Chat finds answers using YOUR organization's
  knowledge — and the public web only if you opted in
        │
        ▼
  The answer comes back to you
        │
        ▼
  A compliance record is kept (that the review ran),
  accessible only to logged, role-gated oversight

At no point does your message leave your organization’s workspace for a public AI, and at no point is a colleague casually reading along.

What you control

Privacy isn’t only something done for you — there are choices that are yours:

What you share. A conversation stays private until you share or branch it. You decide who comes in.
What Chat remembers. Some processing — like the AI remembering useful context about your work — is optional, and you can turn it off in your settings. A small number of activities are essential to the service working at all (for example, the basic processing needed to answer you, and the safety review) and can’t be switched off, but the non-essential ones are your call.
Asking about or deleting your data. You can ask what data is held about you and request its deletion, in line with privacy regulations. Data requests and deletion explains how.

A good habit is to be thoughtful about what you paste in — the same common sense you’d apply to any work tool. What not to share has practical guidance.

In short

Your conversations are private to you and your organization by default.
Other companies cannot see your data; the separation is enforced on every request.
Inside your organization, conversations are yours unless you share them; only administrator-granted, read-only oversight roles can review activity for compliance, and that review is logged.
“Monitored” means an automated safety review plus a compliance record — not a person watching you type.
Your data is never used to train public AI and never handed to a public chatbot service.
You control what you share, what Chat remembers, and your right to ask about or delete your data.

Related reading

Safe at work — why Chat is trustworthy for real business use
How Chat is different from other chatbots — the contrast with consumer apps
What not to share — practical guidance on sensitive content
How your data stays isolated — the wall between organizations
Data requests and deletion — your data-access rights