Can Small-Firm Lawyers Use AI Without Compromising Client Confidentiality?

I have had conversations recently with lawyers at small firms about how they’re using AI and have noticed that they’re facing a common conundrum. They’re finding tools like ChatGPT and Claude more useful for legal work and relatively inexpensive, but they’re not sure they can upload client documents or information without violating their duty of confidentiality. Can this be done?

The draw to these less expensive and more versatile forms of AI, rather than tools like Westlaw’s CoCounsel or Lexis+ with Protégé, is that they’re more flexible and are becoming more powerful for legal work.

Connecting Claude or ChatGPT with A2AJ’s free database of Canadian case law can make either of them an immensely useful research tool (as I’ve written about here), comparable in many respects to the output of a product like CoCounsel.

And with features like Claude Code or ChatGPT Work, which allow you to work on Word docs and other files on your computer, lawyers in criminal, family, or immigration law can use AI to do many things more efficiently, like reviewing disclosure or an opposing party’s submissions, or revising a draft factum.

The problem is the uncertainty surrounding the level of privacy for the data you upload to these platforms, and whether a given plan can satisfy duties of confidentiality to the client at a cost that makes sense.

Short answer: lawyers may be able to meet their duty of confidentiality by using affordable business plans offered by OpenAI and Anthropic and by obtaining informed client consent.

Briefly, what are the duties?

They’re comparable across Canada, but to single out two provinces, under rule 3.3-1 of the BC Code of Professional Conduct, lawyers must hold in strict confidence all information concerning a client’s business and affairs acquired in the professional relationship, subject to limited exceptions, including client authorization. Ontario’s Rules of Professional Conduct impose substantially the same obligation under rule 3.3-1.

Both law societies have issued guidance on AI cautioning lawyers about uploading confidential information to AI platforms (BC, Ont). They flag the concern that the information may be retained, accessed by company officials, or used to train AI models.

When it comes to retention, however, BC appears to be a special case. Law Society Rules 10-3 and 10-4 require that a storage provider maintain a lawyer’s records securely, without accessing them “except as is necessary to provide the service,” and that the provider “destroy the records completely and permanently on instructions from the lawyer.”

(Notice, however, that the rule doesn’t require immediate destruction when a lawyer asks, a point I’ll come back to.)

No such thing as a free lunch

The free and lowest-tier paid subscriptions to ChatGPT or Claude (ChatGPT Plus or Claude Pro, each about 30$ a month) would not appear to provide adequate protection for client information in BC or elsewhere in Canada.

Both plans, ChatGPT Plus and Claude Pro, allow users to disable the use of their conversations for model training. But neither guarantees that your data will be accessed only for limited purposes, such as maintenance and security. (They keep open the possibility of each company making use of your data for advertising purposes.)

The alternative is to use either model on a business plan, which offers more privacy through a data processing agreement.

ChatGPT Business, for example, would run roughly $60, with Claude Team offering comparable pricing. (Both are roughly $30 a month but require a minimum of two seats.)

OpenAI’s terms limit access to authorized staff who need it for engineering support, abuse investigations or legal compliance, along with outside contractors who review content for misuse. Deleted chats are removed within 30 days unless required by law to be kept longer (same with uploaded docs you delete), and, by default, your data isn’t used for training. Anthropic’s policy is similar, but it will keep conversations for up to two years if its automated systems flag them as violating its usage policy.

Is a chat a ‘record’ within the meaning of BC’s Rule 10-3?

BC’s guidance on AI suggests that at least some chats could be records. A BC lawyer using ChatGPT or Claude, even on these business plans, might have difficulty complying with the rule — in some cases — because, where a chat (or uploaded doc) does constitute a record, they can’t be sure OpenAI or Anthropic will destroy it completely on their instructions (or within a reasonable period, like 30 days). And I don’t believe any of the more expensive plans from either company would give a lawyer this kind of control.

With this qualification in mind, lawyers in Canada may be able to use ChatGPT or Claude on the business plans described here in compliance with their duties, so long as they disclose the security risks to their client.

The Case for Client Consent

To be clear, law societies (including in BC) do not require that lawyers use only services that delete client information immediately. For a long time now, we have relied on third-party cloud services to store confidential documents, with adequate safeguards, such as encryption in transit and while stored.

So one possibility is to obtain a client’s informed consent to use ChatGPT Business or a Claude Team account, after disclosing the risks involved, including for solicitor-client privilege.

This would mean explaining, in accessible terms, what information would be uploaded and stored (and thus shared with an AI provider), the extent of the guarantees (as described above), and what risks remain.

But it also means explaining the potential benefits.

A lawyer might tell their client that using AI to review disclosure, organize evidence, or assist with drafting could reduce the time spent on preparation and improve the quality of the work. The client might consider these benefits worth the limited risks entailed in using a reasonably secure AI service.

Of course, consent wouldn’t relieve a lawyer of the duty to exercise reasonable care in using the service; i.e., making sure chats are deleted when longer needed and being attentive to notices from the provider about changes to its retention policy.

But with reasonable safeguards and informed consent, there does appear to be a viable path for small-firm lawyers who want to make use of powerful AI tools affordably and without breaching their professional obligations.