AI At Work: Six Guides For Using AI Safely
Six plain-English guides for staff on using AI at work, covering what you can paste, what Copilot can see, how to check an answer and what your policy requires.
Six plain-English guides on using AI at work: what you can paste, what Copilot can see, how to check an answer, how to stop training on your conversations, what to do about meeting notetakers, and what your company policy actually requires.
Guides In AI At Work
What You Can Paste Into ChatGPT At Work: 2 Checks
Your sign-in account and your company policy settle what is safe to paste, and passwords stay out no matter what.
Copilot Can See Your Files: 4 Things To Check First
Copilot does not break permissions. It makes years of old sharing findable in one sentence. Four checks to run first.
How To Tell When An AI Answer Is Wrong: 3 Fast Checks
Fluency is not accuracy. Thirty seconds of checking catches most of it, and you already know how.
AI Training On Your Data: 3 Settings To Turn Off Today
Verified opt-out steps for ChatGPT, Claude and Gemini, and the one that costs you your chat history.
AI Notetakers In Meetings: 4 Checks Before You Accept
Letting a bot into the call also hands the conversation to an outside company. Check who keeps the transcript first.
Your Company AI Policy Explained: 3 Rules That Matter
Approved tool. No client data. Verify before it leaves. Nine words that cover most real situations.
Why We Added This Category
Our help desk started getting a new kind of question last year. Not "how do I fix this", but "am I allowed to do this".
Can I paste a client email into ChatGPT. Why can Copilot see that folder. Someone brought a notetaker to my meeting, is that fine. These are reasonable questions and nobody had written down the answers.
So we did. Six guides on using AI at work, written for the person doing the work rather than the person signing the policy.
What These Guides Assume
They assume you are going to use these tools, because you are, and pretending otherwise just moves the activity somewhere we cannot see it.
They also assume you would rather not cause a problem. In our experience that is almost always true. Most AI incidents we see are not carelessness. They come from somebody who did not know that the free tier behaves differently from the paid one.
The Short Version Of All Six
If you read nothing else on this page, these are the points that do the most work.
- Credentials never go into a chat window. Passwords, MFA codes, API keys. That rule has no exceptions and no tiers.
- Personal sign-in or business sign-in: that single difference decides the rest. Personal tiers train on your conversations by default. Business tiers do not.
- Copilot only shows you what you could already open. When it surprises you, the permission was already wrong.
- Fluent does not mean correct. Check names, numbers, dates and citations before anything leaves.
- Ask before, not after. A question costs two minutes. An incident costs considerably more.
What Changed In 2025 And 2026
This area moves faster than most. The biggest shift was that consumer tiers across the major vendors moved to training on your conversations by default, including ones that previously did not.
That matters because a lot of advice written before 2025 is now wrong in exactly the direction that gets people into trouble. We re-check the vendor claims in these guides every quarter and date them, so you can see how current they are. Government guidance is moving at the same pace, including CISA on the secure adoption of artificial intelligence.
If You Are The One Setting The Rules
These six guides answer "what am I allowed to do today". They deliberately do not answer "what policy should we have", because that is a different job with different stakes.
For that, we keep a separate guide for owners and partners covering exposure, professional obligations and a usable policy template. You will find it at AI guardrails for attorneys, aimed at the partner or owner who signs off on the rules.
Still Not Sure?
Ask us. Using AI at work is exactly the sort of thing where a two minute question beats a careful guess, and our help desk would much rather hear from you before than after.
You can also browse the rest of the knowledge base, or start with getting help if you are not sure who to ask.
New To AI At Work? Start With One Question.
Not sure which of these guides fits what you are about to do? Tell us the tool and the task, and we will point you to the right one. Using AI at work safely starts with asking before, not after.