Is ChatGPT Safe for Lawyers? What Ontario Practitioners Need to Know

It depends on what you put in it. For general legal questions, drafting public content, or brainstorming, ChatGPT is a useful tool. The moment you paste a client's confidential information into consumer ChatGPT, you take on real confidentiality risk under the LSO rules, plus exposure under the US CLOUD Act.

The short answer

Safe is the wrong word to use in absolute terms. ChatGPT is safe for some tasks and risky for others. The line is confidentiality: is the text you are typing something a client would expect you to keep private?

Green light (generally fine):

• Researching a general legal concept or a statute
• Drafting a template you would happily publish, with no client facts in it
• Rewording a blog post, a bio, or marketing copy
• Explaining a procedure to yourself in plain language
• Brainstorming arguments in the abstract, with names and identifying facts stripped out

Red light (real risk with consumer ChatGPT):

• Pasting a client's contract, will, or agreement to summarize it
• A client name combined with the nature of their matter
• Details from a family, immigration, or estate file
• Financial records, medical notes, or anything privileged
• Anything you would not email to a stranger

The difference is not the tool being clever. It is where your text goes, who can reach it, and what happens to it after you hit enter.

Where the risk comes from

Four things stack up when confidential content enters a US consumer AI tool.

LSO Rule 3.3-1 - strict confidentiality. The rule requires you to hold in strict confidence all information concerning a client's business and affairs. That obligation follows the data wherever it goes. If you cannot say where your client's information is stored or who can access it, you cannot say you are meeting the rule. Our compliance page breaks down what "strict confidence" means for a cloud tool.

LSO Rule 3.1-2 - competence. Competence now includes understanding the technology you use. You do not need to be an engineer, but you are expected to know, at a basic level, where client data goes and who controls it. "I did not realize it was hosted in the US" is not a strong answer to a complaint.

US hosting and the CLOUD Act. ChatGPT's consumer and Team products are run by OpenAI, a US company, on US servers. Under the US CLOUD Act, American authorities can compel a US company to produce data it holds, regardless of where the user sits. Your client's information could, in principle, be reached by a foreign government without your client's knowledge. For a fuller treatment of hosting and the CLOUD Act, see the compliance page.

Model-training defaults. Consumer ChatGPT has historically used conversations to improve models unless you turn that off in settings. Controls exist, and you should use them, but the default posture and the US hosting are the underlying concern. Turning off training does not move the data to Canada, and it does not give you a signed agreement you can show the Law Society.

None of this makes ChatGPT malicious. It makes it the wrong container for privileged material.

How to use AI safely as an Ontario lawyer

You do not have to give up AI. You have to be deliberate about what goes where.

1. Never paste identifiable client information into consumer tools. Strip names, addresses, file numbers, and distinguishing facts before you use a general chatbot. If stripping it makes the task pointless, that is a sign you need a different tool.

2. For client work, use a tool built for confidentiality. That means Canadian hosting, a Data Processing Agreement, and a clear promise that your data is never used to train models. These are the same requirements whether the tool is a chatbot or a full platform. See how they map to real features on our related roundup of legal software.

3. Keep an audit trail. You should be able to answer, later, who accessed a client's information and when. Consumer ChatGPT gives you no such record.

4. Separate matters. Client A's information should not bleed into your work on Client B. Per-matter isolation is a basic expectation for anything holding privileged content.

5. Write down your policy. A short internal note on what staff may and may not put into which tools protects the firm and satisfies the competence expectation. For a wider view of doing this well in Canada, read our guide to AI for lawyers in Canada.

The compliant alternative: Surface

Surface gives you the same ask-in-plain-English experience you like about ChatGPT, but over your own files, and built for legal confidentiality.

You ask a question in plain language and get an answer with exact-paragraph citations and a source panel, so you can verify every point against the original document rather than trusting a summary.

The difference is in how it handles your data:

• Canadian hosting only. Your documents are stored on Canadian servers, AWS in Montreal (ca-central-1). Never Azure, never US regions.
• Full documents never leave. At query time only a few short excerpts, typically two to five passages, go to the AI providers (Anthropic Claude and Cohere). Your complete files are never sent.
• No training on your data. Your content is never used to train AI models.
• A DPA available on request. Not something you have to negotiate, and not reserved for enterprise buyers.
• Audit logs and per-matter isolation. Access is recorded, and matters stay separated.

Surface connects to Gmail, Outlook, Google Drive, and OneDrive, so the documents and email you already have become searchable without you moving anything around. It runs as a desktop app on macOS and Windows.

For the fuller picture of how AI fits into a compliant Canadian practice, our AI for lawyers in Canada guide walks through it.

Why this matters

The risk here is not dramatic. No one is going to fine you the day you paste a contract into a chatbot. The risk is quieter: a confidentiality obligation you cannot demonstrate you met, sitting behind a US company's servers, with no agreement in your file and no record of who saw what.

A tool that keeps data in Canada, never trains on it, gives you a DPA available on request, and shows you a verifiable citation for every answer removes that whole category of worry. That is the entire reason to choose one over the other for client work. The compliance page lays out each requirement and how it is met.

This article is general information, not legal advice. For your own obligations, consult the Law Society of Ontario.

FAQ

Can lawyers use ChatGPT?

Yes, for the right tasks. General research, public drafting, and brainstorming without client facts are fine. The caution applies to confidential client information, which should not go into consumer ChatGPT.

Is it a confidentiality breach to use ChatGPT?

Not by itself. It becomes a risk when you enter identifiable or privileged client information, because that data is processed on US servers you do not control, which is hard to square with LSO Rule 3.3-1.

Does ChatGPT train on my data?

Consumer ChatGPT has historically used conversations to improve models unless you turn that setting off. You can disable it, but that does not change where the data is hosted or give you a signed agreement to rely on.

Is ChatGPT hosted in Canada?

No. OpenAI is a US company and its consumer and Team products run on US servers, which brings US CLOUD Act exposure. If Canadian hosting matters for your files, you need a tool that offers it directly, such as Surface on AWS Montreal.

What can I safely use ChatGPT for as a lawyer?

Anything with no confidential client content: general legal concepts, statute overviews, template drafting, marketing copy, and abstract brainstorming. For work involving real client facts, use a tool with Canadian hosting, a DPA available on request, and no training on your data.

Get Surface

If you handle client documents, the safe move is a tool built for confidentiality from the start. Surface is a desktop app for macOS and Windows, $49/month CAD flat, with a 14-day free trial and no credit card required.

Download Surface and start your free trial.