Usually the AI company is not how this reaches your manager. The device, the network and the account are. Which of those you used decides the answer, so it is worth taking them one at a time.
Work account on a business or enterprise plan
Assume yes. Business and enterprise tiers exist partly so that an organisation retains control of what its people put into the tool — that is the feature being sold. Administrators can typically reach conversation history, export it, and retain it under company policy.
This is not a loophole or a betrayal; it is the deal. If your employer pays for the seat, the account is theirs. Treat anything typed there as you would treat an email sent from your work address.
Personal account, work laptop
Here the AI provider gives your employer nothing. The device might give them a great deal.
- •Managed devices commonly run endpoint monitoring. Depending on configuration that can include screenshots, keystrokes, visited URLs, or clipboard contents — none of which cares which website you were on.
- •Browser profiles sync. A work-managed browser can report extensions, history and sometimes more back to an administrator.
- •Installed software can be inventoried. Including which AI apps you have.
The honest summary: on a laptop you did not buy, assume anything on screen can be seen. That is true of AI chats, personal email, and everything else.
Personal account, work network
Weaker, but not nothing. Traffic is encrypted in transit, so a network administrator generally cannot read the words. They can usually see which sites a device connected to and when — and on a network doing TLS inspection with a corporate certificate installed on the device, more than that.
So the realistic risk on your own phone on office wifi is that someone can tell you used an AI assistant at 3pm, not what you asked it.
Personal account, personal device, personal network
Your employer has no route to it. What remains is the ordinary exposure everyone has: the provider stores the conversation, staff and subprocessors can access it for abuse review and support, and it can be produced under legal process. Those are real, and they are not your manager.
Three things people get wrong
- •Incognito does not help. It stops your browser writing local history. It is invisible to monitoring software and to the network. It was never a privacy mode in the sense people use the phrase.
- •Deleting the chat afterwards does not undo monitoring. If something captured the screen while you typed, deleting the conversation later removes the provider's copy and not that one.
- •A personal login on a work device is not a personal session. The account is yours; the machine watching you type is not.
What to do about it
Roughly in order of how much it buys you:
- •Use your own device for anything personal. This single change resolves most of the above and no software substitutes for it.
- •Keep work questions on the work account. Partly compliance, mostly that mixing them is how personal things end up somewhere discoverable.
- •Read the acceptable-use policy once. Many organisations now say plainly what is monitored. It is usually more than people assume and less than they fear.
- •Leave the identifying detail out. Useful everywhere, and the only measure that also covers the provider.
Where Secure AI fits, and where it does not
Secure AI removes identifying details before a message leaves your device, so the AI provider never learns whose question it is. On your own phone that closes the biggest remaining gap.
It does not defeat device monitoring, and we are not going to pretend otherwise. If your employer's software is capturing your screen, no assistant — ours included — changes what it captures. The answer to that problem is a different laptop, not a different app.
Try Secure AI free
Frontier AI chat — encrypted by default, with your identity stripped before any request reaches the provider.