Can you put client information into ChatGPT in Canada?

Staff are already pasting client emails, meeting notes and spreadsheets into AI tools, usually on a personal account and usually because it saves them an hour. The question for an owner or manager is whether that is allowed, and what has to be true for it to be.

This page explains what the federal privacy law and Canada's privacy regulators ask of a business using generative AI, plus the Nova Scotia and Quebec rules that catch some organizations. It is general information, not legal advice. For health, legal or financial records, talk to a privacy lawyer about your specific setup.

Does it matter which ChatGPT plan you use?

Yes, more than anything else on this page. Consumer accounts and business accounts handle your data differently.

• ChatGPT Free, Plus and Pro: OpenAI says data sharing for model training is on by default on personal accounts, and users can turn it off.
• ChatGPT Business and Enterprise: OpenAI says that by default it does not use inputs and outputs from these plans to train its models.
• Claude: since August 2025, Free, Pro and Max users choose whether their chats can be used for training. Anthropic says this does not apply to Claude Team, Enterprise or the API, which fall under its commercial terms.
• Microsoft Copilot Chat: when staff sign in with a work account on an eligible Microsoft 365 plan, prompts and responses are covered by Microsoft's enterprise data protection, and Microsoft says they are not used to train the underlying models.

A business plan also gives you an admin console, the ability to remove a departing employee's access, and a data processing agreement. A staff member's personal account gives you none of that. If your team uses AI at all, the first fix is to move them onto a business account in your organization's name.

What does PIPEDA require when you use AI?

PIPEDA, the federal law covering most private-sector businesses in Nova Scotia and other provinces without their own private-sector law, applies the same way whether a person or an AI tool handles the information. Three parts matter most here.

• Purpose and consent. You can use personal information for the purposes you identified when you collected it, or ones a reasonable person would expect. If clients gave you their details to do their taxes, using them to train a marketing assistant is a new purpose.
• Accountability. PIPEDA's accountability principle says you remain responsible for personal information you transfer to a third party for processing, and you must use contracts or other means to protect it. The Privacy Commissioner's guidelines on processing across borders apply the same rule when that vendor stores data outside Canada.
• Safeguards. Protection has to match the sensitivity of the information. Client financial or health records need more than a shared login.

What do Canada's privacy commissioners say about generative AI?

In December 2023, the federal, provincial and territorial privacy regulators jointly published principles for generative AI. Most of the document is aimed at developers, but several points are written for organizations using these tools:

• Where possible and reasonable, use anonymized or de-identified information in prompts.
• Only enter personal information, especially sensitive information, where you are authorized to.
• Tell affected people when a generative AI tool will be used.
• Take reasonable steps to check that outputs are accurate for the purpose.
• Accountability for decisions stays with the organization, not the tool.
• Consider a privacy impact assessment, and set retention schedules for prompts and outputs.

These principles are guidance, not law. They do show how the regulators read the existing laws, which makes them a sensible baseline for your own policy.

What if you work for a Nova Scotia public body?

Different rules apply. Nova Scotia's Personal Information International Disclosure Protection Act (PIIDPA) covers public bodies, municipalities, and service providers acting on their behalf. It requires personal information to stay in Canada, and be accessed and disclosed only in Canada, unless an exception applies. One exception allows storage or access outside Canada when the head of the public body decides it meets the necessary requirements of operations, and that decision must be reported to the Minister of Justice.

Most AI tools process data outside Canada, at least some of the time. Microsoft, for example, says Copilot Chat routes requests to the nearest data centres but can use other regions when demand is high. If your nonprofit or firm delivers services under a provincial or municipal contract, ask the public body before putting any of that work's personal information into an AI tool.

Does Quebec's Law 25 apply to you?

It can, if you handle personal information about people in Quebec. Quebec's private-sector privacy act, as amended by Law 25, adds two rules that matter for AI. You must tell a person when a decision about them is based exclusively on automated processing (section 12.1). And before personal information leaves Quebec, including to a cloud AI service, you must assess whether it will be adequately protected (section 17). Have the section text checked by a lawyer if you have Quebec clients, since secondary summaries of Law 25 vary.

A worked example

Example: a 12-person accounting firm in Dartmouth finds that two staff have been pasting client emails into personal ChatGPT accounts during tax season to draft replies. Some of those emails include SINs and income figures.

A sensible response:

1. Move everyone to one business AI account in the firm's name, with training off by default, and close the personal-account habit.
2. Set a rule: no SINs, account numbers or full financial statements in any prompt. Replace them with placeholders such as "Client A" before pasting.
3. Add a line to the engagement letter and privacy notice saying the firm uses AI tools from named vendors to draft correspondence, with staff review.
4. Require a human to read every AI draft before it goes to a client.
5. Keep a short record of which tool is approved, what data may go into it, and who reviewed the setup.

A simple data rule for staff

Most small teams do well with three categories, written on one page.

Category	Examples	Rule
Fine to use	Public website text, your own templates, general questions, internal process notes with no names	Any approved business AI tool
De-identify first	Client emails, meeting notes, case summaries	Remove names, addresses, account numbers and other identifiers, then use an approved business tool only
Never	SINs, health records, banking details, legal case files, anything covered by a contract with a public body	Keep out of AI tools unless a lawyer or privacy officer has approved a specific setup

Health information custodians have their own provincial law to follow. In Nova Scotia that is the Personal Health Information Act. The "never" row is the safe default until someone has checked your setup against it.

What to do next

• Find out which AI tools staff already use, and on which accounts.
• Choose one approved business tool and close personal-account use for work.
• Write the three-category data rule and share it with staff.
• Update your privacy notice to say you use AI tools and how.
• Check contracts with any public body for data location rules.
• Name one person responsible for the AI policy and review it every six months.

Our page on using AI responsibly in a nonprofit covers a similar policy for charities. If you want help setting up an approved tool and a staff policy together, that is part of our AI adoption and training work, and our AI workspace service covers setups with approval steps built in.