New:The AI Integration Day: one day at your office to put AI to work.The AI Integration Day

Discover

Can you put confidential data into ChatGPT?

It depends on the account, the version of the tool and what you paste into it. Here's where your data goes when your team uses an AI assistant, what you should never put in it and how to frame its use without banning AI.

Peut-on mettre des données confidentielles dans ChatGPT?

Monday morning. Someone on the team pastes a supplier contract into an AI assistant to get a summary. 30 seconds later, the summary is clear, well done, ready for the meeting.

Then a small question comes up, a little late: where did that contract go?

That's not a reason to ban AI. Banning it often pushes the team to use it anyway, with a personal account and no rules. It's a reason to know where your data goes and to set a few simple rules. Here's the essentials, without jargon.

Where does your data go when you paste it into an AI assistant?

It leaves your computer and is processed on the servers of the company that provides the tool. What happens next depends on the account you use and its settings.

With a personal or free account, depending on the tool and the settings, your conversations may be kept for a while, reviewed for security reasons or used to improve the models. The business versions of Claude, ChatGPT, Copilot and Gemini generally commit not to train their models on your data, and they give the company control over the accounts.

The terms change often. Before choosing, read the ones for the version you pay for, or ask someone to go through them with you.

A detail people forget: with a personal account, the history belongs to the person. The day they leave the company, your summarized contracts leave with them.

The same tool, 2 accounts: it's not the same thing for your data. Always check the terms of your version.

What should you never put into an AI assistant?

Never passwords or access codes, whatever the account. And never personal information or confidential information in a personal or free account.

Concretely: no social insurance numbers, credit card numbers, health records, customer contact details, salaries, contracts under a confidentiality agreement or prices that give you your edge.

One simple habit helps a lot: remove what identifies people. To rephrase a delicate email, translate a letter or summarize a complaint, the assistant doesn't need the customer's name, address or phone number. "Customer A" works just fine.

What about grey areas? A quote for a customer, a sales table, an inspection report: in a business account, with the right settings, it's generally fine. In a personal account, no. When in doubt, ask the person responsible before pasting.

And many uses carry little risk, even with a free account:

Examples

  • Rephrasing a text that will be public, like a job posting
  • Preparing the outline of a presentation or a training session
  • Summarizing a complaint after removing the customer's name
  • Writing an email with "Customer A" instead of the real name
To rephrase an email, the assistant doesn't need to know who the customer is.

What does Law 25 say about AI and personal information?

Law 25 doesn't name any AI tools. It governs what your company does with personal information, whatever the tool. In Quebec, it modernized the rules on the protection of personal information in the private sector.

Several assistants process data outside Quebec. Putting personal information into them therefore deserves some thought before, not after. This isn't legal advice: for your situation, talk to your legal advisor or consult the Commission d'accès à l'information du Québec.

Among what the law provides, 4 points directly affect the use of AI:

  1. A person in charge

    Every company has a person in charge of protecting personal information. By default, it's the person with the highest authority in the company. They can delegate this role in writing.

  2. An assessment before leaving Quebec

    Before communicating personal information outside Quebec, the company must assess the privacy risks.

  3. An automated decision must be disclosed

    When a decision about a person is based solely on automated processing, the company must inform them.

  4. A contract with the provider

    When the company entrusts personal information to a provider to process on its behalf, it must set out in writing the measures that protect it.

How do you frame the use of AI on your team?

With 3 things: simple rules, business accounts and well-managed access. No need for a 40-page manual.

One page is enough. What you can do with AI, what never goes into it and who to call when in doubt. A clear rule beats a ban nobody follows.

Next, switch to business accounts. The data stays under your control, you choose who has access to what and access is removed the day someone leaves the company. If you already use Microsoft 365 or Google Workspace, first check what your plan includes: Copilot and Gemini have business versions there, sometimes already included.

Finally, train the team. Showing how to remove a name or summarize without copying the whole file takes a few minutes and avoids a lot of trouble. It's one of the topics we cover when we train a team, for example during the AI integration day.

One page, 3 sections: that's enough for everyone to know what to do.

What changes when AI is connected to your systems?

The question changes: you no longer ask what you paste into the assistant, but what it is allowed to see and do.

An assistant connected to your emails, your customer files and your management software can read much more than a contract pasted by hand. That's what makes it useful. It's also why it must be framed from the start. And an AI agent, which chains tasks on its own, needs even more attention: you decide in advance what it can do without asking, what waits for approval and what it never does.

3 rules we apply every time. The assistant signs in as each person and sees only what that person is allowed to see: your access, nothing more. Anything that commits the company, like sending to a customer, paying or deleting, goes through a person. And every action is logged, so you know who did what.

We also look at where the data is processed and stored, especially when it contains personal information. See how that works in the assistant connected to the tools of a professional services firm.

The assistant has the person's access, nothing more. Anything that commits the company waits for their approval.

Where do you start this week?

With an honest picture of what's already happening. Ask the team which AI tools they use, without judging: you want the truth, not hiding. The goal isn't to monitor the team. It's to give them a framework to use AI without second-guessing every time.

Then, a few simple steps are enough to get off on the right foot. And review your rules every year: the tools change fast, and so do their terms.

Examples

  • List the AI tools already in use, and with which accounts
  • Switch to business accounts for the tools you keep
  • Write the rules page and present it to the team
  • Name the person who answers questions and follows changes to the tools

Frequently asked questions.

Is it safe to put business data into ChatGPT?

It depends on the account. With a personal or free account, depending on the settings, your conversations may be kept or used to improve the models. Business versions generally commit not to use your data for training and give the company control over the accounts. Read the terms of the version you use.

What information should you never put into an AI assistant?

Never passwords or access codes, whatever the account. In a personal or free account, also avoid personal information (social insurance numbers, health records, customer contact details), salaries, confidential contracts and prices that give you your edge.

Does Law 25 apply to the use of AI?

Yes, as soon as personal information is involved, whatever the tool. Among other things, it provides for a person in charge of protecting that information within the company and a risk assessment before communicating it outside Quebec. For your situation, consult a legal advisor.

How do you frame the use of AI in an SMB?

With a page of simple rules (what you can do, what never goes in, who to call when in doubt), business accounts controlled by the organization and a short training session for the team. A clear rule beats a ban nobody follows.

Can AI connected to our systems see everything?

It shouldn't. A properly configured assistant signs in as each person and sees only what that person is allowed to see. Anything that commits the company, like sending something to a customer or making a payment, goes through a person. Every action is logged.

AI and confidentiality aren't in conflict. The danger is use without rules, with accounts the company doesn't control and data whose destination nobody knows.

A few clear rules, the right accounts and well-thought-out access: your team keeps the speed of AI, and your data stays yours.

Read next

An idea in mind?

Let’s chat!

By submitting this form, I agree to the privacy policy.