AI for your firm: Part 6 of 8. A short series for law firm leaders on getting AI right, from the risks you can’t see to the policy, tools and plan that fix them.
Quick answer: Secure AI tools for law firms don’t train on your data, tell you where it’s processed and stored, let you control retention and deletion, keep audit logs, support SSO and multi-factor authentication, put security commitments in the contract, help you verify output, and fit your regulator’s guidance. Get every answer in writing before client information goes in.
A vendor tells you their AI is “enterprise-grade” and “bank-level secure.” Your partner asks one question: “But is it safe to put client files in it?”
You can’t answer from a sales page. Neither can anyone else. What you can do is ask the same set of questions every time, get the answers in writing, and compare them against what your regulator expects.
This guide gives you the criteria and the questions. It’s vendor-neutral: the same checks apply to a general-purpose assistant, a legal-specific tool, or an AI feature inside software you already use.
Our background: we’re a certified Clio consultant, a LEAP certified consultant and a certified Actionstep Legal Financial Management consultant, and we’re an Actionstep partner. We don’t sell AI software, and we hold every tool to the same checklist, whichever vendor it comes from.
Start with what you’re protecting
Before you compare tools, decide what might go in them. The answer changes how strict you need to be.
- Tier 1: Public or general information. Brainstorming, marketing copy. Low risk.
- Tier 2: Internal firm information. Templates, policies. Moderate risk.
- Tier 3: Client confidential information. Anything from a client file. This is where the criteria below matter most.
ABA Formal Opinion 512 says lawyers must assess the likelihood of disclosure, the sensitivity of the information and the difficulty of implementing safeguards. The Law Society of BC’s guidance says ideally client confidential information, including anything identifying the client, would be left out of what you give a generative AI tool. If you can’t leave it out, a tool built to protect it matters even more.
Also note what regulators can’t do. The Law Society of BC says it cannot endorse any product or tool. The vetting is yours. Document it.
What makes secure AI tools for law firms? Eight criteria
1. Does it train on your data?
This is the first question. The ABA opinion says some tools are “self-learning,” and that this raises the risk that one client’s information could be disclosed improperly, even within one firm. It says informed consent is required before you input client information into such a tool, and that boilerplate language in engagement letters isn’t enough.
Look for a written statement that your inputs and outputs aren’t used to train models for anyone else, on every plan you’d use.
2. Where is the data processed and stored?
Data residency matters to some firms and clients, not others. We haven’t found a blanket regulator rule setting a location for all firms, so check your law society or Bar Association and your client contracts. Ask the vendor where data is processed and stored, whether you can choose a region (including Canada), and who their subprocessors are.
3. Confidentiality, retention and deletion
How long does the vendor keep prompts and files? Can you delete them? Do their staff review conversations? The Office of the Privacy Commissioner of Canada says prompts shouldn’t be retained, used for secondary purposes or disclosed unless otherwise required. ABA 512 adds: find out whether the tool keeps information after you stop using the service or claims proprietary rights to it. Get the answer in the contract, not just the FAQ.
4. Audit logs and admin controls
If something goes wrong, can you see who used the tool, when, and with what? Can administrators manage users, turn features off and export logs? Law Society of BC Rule 10-4 requires lawyers to notify the Executive Director immediately if they believe there’s been improper access to their records. That’s hard to do if you can’t see what happened.
5. Access controls: SSO, MFA and permissions
Look for single sign-on (SSO), multi-factor authentication and role-based permissions. With SSO, when someone leaves, one account switched off removes their access to the AI tool too.
6. Vendor terms and security assurances
ABA 512 says all lawyers should read and understand a tool’s terms of use, privacy policy and related contract terms, or have someone who has read and analyzed them do it. Lawyers may need to consult IT or cybersecurity experts. Key points include:
- Is the confidentiality commitment enforceable in the contract?
- Will you be notified of a breach, or if someone demands your data?
- What are the limits on the vendor’s liability?
- What happens to your data if the vendor is acquired or shuts down?
- What independent security reports or certifications can they share?
7. Accuracy and verification features
No tool eliminates the need to check, but some make checking easier by showing sources and linking to the underlying documents. The Office of the Privacy Commissioner says organizations should evaluate a tool’s validity and reliability for its intended purpose. Test before you decide.
8. Fit with your regulator
Match the tool to the guidance that applies to you. In British Columbia, the Law Society says AI products need the same thoroughness as other technology, including Rules 10-3 (Records) and 10-4 (Security of records). Rule 10-3(4) covers storage providers, who must not access or copy records except as needed to provide the service. In the US, start with ABA 512 and your state Bar Association’s guidance. Court rules matter too.
Four types of AI tools, compared fairly
Each type has trade-offs. None is automatically the safest.
| Type | Often a good fit when | Watch for |
|---|---|---|
| General-purpose AI assistant | Tier 1 tasks, brainstorming | Differences between consumer and business terms |
| Legal-specific AI tool | Research, drafting, document review | Retention and what it does with your files |
| AI inside software you already use | You want one set of controls and one contract | Which features are on by default |
| Firm-hosted or private setup | High-sensitivity work; strict client requirements | Cost, upkeep and security are your responsibility |
The Law Society of BC notes a private tool that only uses information provided by the firm might help avoid some confidentiality concerns. Sometimes the best option is a feature you already own, with the right settings.
🧠Not sure which kind of tool fits your firm?
Our free AI readiness assessment lists the AI features you already own and shows where your confidentiality gaps are.
The ask-the-vendor checklist
Copy these into an email or procurement form. Ask for written answers.
Data use
- Do you use our inputs or outputs to train or improve any model? Does that apply to all plans?
- Does any person at your company, or any third party, review our prompts or files? Under what conditions?
Location and retention
- Where is our data processed and stored? Can we choose the region?
- Who are your subprocessors, and where are they?
- How long is data retained? Can we delete it, and can you confirm deletion in writing?
- What happens to our data if we cancel, or if you’re acquired?
Access and monitoring
- Do you support SSO and multi-factor authentication?
- Can we assign roles and restrict features by user or group?
- Do you keep audit logs of user activity? Can we export them? For how long?
Security and incidents
- What independent security reports or certifications can you share?
- How and how quickly will you notify us of a breach or a legal demand for our data?
- Which of your security commitments are in the contract, not just on your website?
Terms and accuracy
- What are your limits on liability, and what indemnities do you offer?
- How does the tool show its sources, and how can our lawyers verify its output?
Keep the answers. If a vendor won’t answer a question in writing, that’s information too.
Test before you commit, and re-check after
Don’t rely on a demo. Run a short trial with non-confidential or anonymized documents (our 90-day plan shows how to structure a pilot). Pick a task and a handful of documents you know well, compare the tool’s output to a careful manual result, and check that the settings do what the vendor says. This mirrors an example in ABA 512: a lawyer who has tested a tool on a smaller set of contracts, comparing its summaries to a manual review, would not necessarily have to manually review every document in a later, larger set.
After you buy:
- Configure before rollout. Turn on SSO and MFA, turn off data-sharing options you don’t want, and set retention.
- Write it into policy. Add the tool to an approved-tools list in your law firm AI policy.
- Train your team. ABA 512 says supervisory duties include training lawyers and staff on ethical and practical use.
- Re-check yearly, and whenever the vendor changes its terms.
- Watch for the unapproved alternative. If the approved tool is hard to use, people drift back to personal accounts. See Shadow AI in law firms.
FAQ
Is a legal-specific AI tool automatically safer than a general one?
No. Legal-specific tools may be designed with law firm needs in mind, but you still need to review their terms, retention, hosting and security. Judge every tool by the same checklist.
Does the data need to stay in Canada?
It depends on your regulator, your clients and your contracts. We haven’t found a blanket rule for all firms. Check your law society’s guidance and your engagement terms, and ask vendors where data is processed and stored.
Do we have to tell clients which AI tools we use?
It depends on your jurisdiction and the facts. ABA 512 says informed consent is required before putting client information into certain self-learning tools. The Law Society of BC says it’s prudent to make clients aware of how you plan to use AI. Build your approach into your policy.
Need an independent view on your shortlist?
We’re vendor-neutral consultants. Our free AI readiness assessment shows what you already own, where your confidentiality gaps are and the guidance that applies to your jurisdiction. If you’d like a deeper review of how AI is used in your firm, see does your firm need an AI audit? We work with law firms in every province and territory across Canada and throughout the United States.
Ready to see where you stand? Get your free AI readiness assessment · Explore AI implementation for law firms
This article is general information, not legal, ethics or cybersecurity advice. Consult your law society or Bar Association’s practice advisors and your IT or security professionals about your specific situation.
