← Use AI on Board, M&A and People Matters Without Leaking Them

Free lesson · lesson 1 of 7

What Happens to Material You Give an AI Tool

Where your words go after you press send

When you press send in an AI chat tool, your prompt and any uploaded files leave your laptop. They go to the provider's computers, usually called servers. From there, copies can end up in several places:

  • Chat history: the conversation saved in your account so you can return to it. Anyone who can open that account can read it.
  • Provider logs: records the provider keeps of requests, for example to fix faults or investigate misuse. Under some conditions, provider staff may be able to see them.
  • Backups: recovery copies that may remain for a period after you delete a chat.
  • Model improvement: some services may use prompts and responses to train or improve future versions of their AI model.

Which of these applies depends on the product, the account type and the settings. Vendors also change their terms. So don't assume. Check the current vendor documentation and ask your IT or security lead what applies to the tool you are using.

Example: you delete a chat in which you drafted board minutes. The chat disappears from your screen. That alone does not tell you whether logs or backups still hold a copy.

See itThe journey of a prompt

Your screen shows only the first stop. The later stops depend on the product, account and settings.

Try it

You delete a chat containing draft board minutes. Which statement is accurate?

A small paste with a large reach

Here is how an ordinary shortcut becomes a leak.

It is late. You are finishing an acquisition memo, and three paragraphs read awkwardly. You open the AI chat tool you use at home, which is still logged in on your work laptop. You paste the paragraphs and ask it to tighten the wording.

Those three paragraphs name the target company. They give the price range you are prepared to offer and the date you plan to approach the target's board. All of that has now gone to a service your company has no agreement with. That means no contract sets how long the provider keeps the content, who at the provider can see it, or whether it may be used to improve their models.

The tool did nothing unusual. It did what it does for every user. The problem was where the content went, not how well the tool rewrote it.

You wanted better sentences. What left the building was the deal itself: who, how much and when.

Picture itA late-night shortcut

Look at which account the tool is logged into, not just what the document says.

Try it

In the example, which three pieces of deal information left the company with the pasted paragraphs?

Sensitive is more than the main file

Many leaders guard the board paper and forget everything around it. In an AI session, the matter can leak through many smaller pieces.

Suppose you never upload the paper. You only type: "Make this note to the chair clearer. It covers Project Heron, the $40 million offer, closing in March, and the planned restructuring of the Leeds team."

That one prompt reveals a codename, a figure, a date and a staff matter. Anyone who sees it can reconstruct most of the story.

The pieces to watch:

  • Names of people, companies and advisers
  • Deal codenames, which mean something to anyone who has heard them
  • Figures and dates: prices, salaries, deadlines
  • The wording of your prompt, which often explains the situation more plainly than the document does
  • Attachments and pasted spreadsheet cells, which may include hidden tabs, comments or extra rows
  • The saved conversation, which collects all of the above in one place

Treat each of these as part of the confidential material, not just the file.

Try itWhat each piece gives away

Flip each card and ask whether that piece alone could identify the matter.

Try it

Name three parts of an AI session, other than the main document, that can reveal a confidential matter.

Same brand, different account, different rules

Many AI tools come in at least two versions under the same name:

  • A consumer version: a personal account you sign up for yourself, often free or paid by card.
  • A business version: an account your company arranges, covered by a company agreement and managed by your IT team.

They can look almost identical on screen, but the terms can differ. These include how long content is kept, whether it may be used to improve models, and what controls the company has. The exact differences vary by vendor and change over time, so confirm them with current vendor documentation and your IT or security lead.

The practical trap: you are on your work laptop, the tool's logo is the one IT approved, but you are signed in with your personal email. In that case you are usually outside the approved setup, even though everything looks familiar.

Quick ways to tell which account you are in:

  • Which email address is shown in the account menu?
  • Did IT give you access, or did you sign up yourself?
  • Does the account show your company's name or workspace?
See itPersonal account vs company-approved account

Personal (consumer) account

  • You signed up yourself, often with a personal email
  • Terms are between you and the vendor, not your company
  • Content handling depends on consumer terms and your own settings
  • Usually outside the approved setup, even on a work laptop

Company-approved (business) account

  • Access arranged by IT, signed in with your work identity
  • Covered by an agreement between your company and the vendor
  • Settings managed by the company
  • Inside the approved setup, within the limits IT has set

The logo can match while the terms do not. Check the account, not the brand.

Try it

Your company has approved the business version of an AI tool. On your work laptop you open the same tool, signed in with your personal email. Where are you?

Everything attached to the tool counts

An approved AI tool can still leak if something attached to it sends content elsewhere. Four common add-ons:

  • Connector: a link that lets the AI tool read from or write to another app, such as your email or file storage.
  • Plugin: an add-on that lets the AI tool call another company's service to complete a task.
  • Browser extension: a small program in your web browser. Some can read the pages you open, including a board paper you are viewing.
  • AI meeting note-taker or transcription feature: records or transcribes a video call and stores the result, sometimes with a separate provider.

Each can pass content to a service you did not choose and your company may not have approved.

Example: you hold a sensitive staff discussion on a video call. One participant has an AI note-taker that joins automatically. The full discussion, with names and the issue, is now transcribed and stored by that note-taker's provider.

The rule to remember: a tool is only as contained as everything attached to it. An approved chat tool combined with an unapproved extension is not an approved setup.

Try itMatch each add-on to what it does

Tap one on the left, then what it goes with on the right.

0 of 4 matched

Each of these can move content beyond the tool you thought you were using.

Try it

A board call is about to start, and you notice an AI transcription feature is switched on. What is the main concern?

A pause before you send, and where legal questions go

Put this lesson together into a short pause before any confidential work goes into an AI tool:

  1. Check the account. Is this the company-approved account or a personal one?
  2. Check what is attached. Are any connectors, plugins, extensions or note-takers active?
  3. Check what you are sharing. Look beyond the file to names, codenames, figures, dates, pasted cells and the wording of your prompt.
  4. Confirm if unsure. Check the approved-tools list or ask your IT or security lead.
  5. Then send, or stop.

Later lessons add sensitivity levels, setup choices and redaction to this routine.

One boundary: this course does not give legal advice. Whether a particular disclosure breaches data protection, employment, securities or confidentiality obligations is a question for counsel. If you think material has already gone into an unapproved tool, tell your IT or security lead and counsel promptly. Don't quietly delete the chat and hope. Deleting may not remove other copies, and they need to know so they can act.

Try itPut the pre-send pause in order

Put these in order: tap two to swap them, or use the arrows.

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5

Account first, then attachments, then content, then confirmation, then the decision.

Try it

You realize part of an acquisition memo went into an unapproved tool last week. Who should judge whether this breached legal or regulatory obligations?

In short

Once you press send, your prompt and files can end up in chat history, provider logs and backups, and some services may use them to improve their models. What applies depends on the product, account type and settings, so check current vendor documentation and your IT or security lead. Sensitive content goes beyond the main document: names, codenames, figures, dates, prompt wording, attachments, pasted cells and saved conversations can each reveal the matter. A personal account of an approved brand is usually outside the approved setup, and connectors, plugins, extensions and meeting note-takers can carry content to services you never chose. Legal questions about a disclosure belong with counsel.

Carry on with the whole course

6 more lessons, a tutor you can ask anything in every one, quizzes, flashcards and a certificate. $47.

See everything it covers

Not ready yet? Have this lesson emailed to you.

Email me this free lesson and occasional news about this course. Unsubscribe any time with one click.