A Law Firm's Guide to Evaluating AI Vendors for Client Confidentiality

2026-04-09 · 8 min read · Legal AI · 0 views

Why this guide exists

If you're evaluating AI tools for your law firm, you're not just evaluating software. You're evaluating a business relationship with implications under ABA Model Rule 1.6 (Confidentiality of Information), your state's bar rules, and your firm's malpractice exposure.

The AI vendors who want your business are happy to give you a demo. What fewer vendors give you is a straight answer on where your client data goes, who can see it, and what happens to it after you submit it.

This guide gives you seven questions to ask every AI vendor before you sign anything. It is not legal advice. It is a framework for asking the right questions and knowing what the answers should be.

Question 1: Where does my client data actually go?

The only honest answer is one of three: it stays in your environment, it goes to the vendor's environment (and may be used to train models), or it goes to a third-party environment the vendor doesn't fully control.

If the vendor says "it stays private," ask what that means technically. Private deployment on your own server? Private deployment on a dedicated (not shared) cloud environment? Or "private" meaning the vendor promises not to look at it?

Vague privacy language is a red flag. Specific architecture descriptions are a green flag.

Question 2: Do you use my firm's data to train your models?

Some AI vendors explicitly state that client data submitted through their platform is not used for model training. Others make this opt-out, which means opt-in by default. A growing number have changed their position post-regulation, but not all have updated their contracts.

If you're handling confidential client information, you need this answer in writing, in the contract, not just in a sales presentation.

Question 3: What are your data retention and deletion policies?

If you submit a client matter to an AI tool and then the matter closes, what happens to that data? The vendor may retain it. They may retain it for years. They may share it with subprocessors you've never heard of.

Ask for the specific retention schedule. Ask what happens to data if you cancel your account. Ask whether data is deleted or merely "de-identified." De-identified data is not the same as deleted — and in a world of re-identification research, the distinction matters.

Question 4: Who at your company can access my firm's data?

Most enterprise AI vendors have internal access policies that limit what employees can see. Ask whether your data is accessible to the vendor's engineering or support teams, and under what conditions that access is granted.

The right answer involves technical controls: encryption at rest and in transit, role-based access controls, audit logs of who accessed what and when. If the vendor can't describe their internal access controls, that's a gap.

Question 5: What is your breach response protocol?

This is where many vendor evaluations stall — because the answer is often "we follow industry-standard protocols" or "our terms of service govern that."

You want specifics. What is the notification timeline if your data is exposed? What is the vendor's liability under the contract? Does the vendor carry cyber liability insurance, and if so, what are the coverage limits? Does the vendor have a published security certification (SOC 2, ISO 27001)?

If a vendor resists giving you their incident response summary or security certifications, that resistance itself is information.

Question 6: Can I deploy on my own infrastructure?

Some law firms are not willing to put certain client data into any third-party environment — even a trusted one. If that describes your firm's risk posture, this question eliminates entire categories of vendors.

Self-hosted or private-deployment AI means your firm's IT team controls the infrastructure. The vendor provides the software, the model weights, and the updates — but the data never leaves your server. This is the highest-control option and the one most aligned with strict confidentiality obligations.

Not every firm needs this. But if your practice involves high-sensitivity matters — M&A due diligence, patent applications, client financial information, immigration cases — the question is worth asking.

Question 7: What does your bar ethics compliance support look like?

This is the question most law firms forget to ask. If you're using an AI tool in your practice, you're responsible for ensuring that use complies with your state's bar rules. Some AI vendors have begun publishing formal ethics opinions, bar association filings, or legal ethics guidance documents that address how their specific deployment architecture fits within Rule 1.6 obligations.

If a vendor has this documentation, read it. If they don't, that doesn't disqualify them — but it means you should do your own analysis before relying on the tool for client matters.

What to do with the answers

Once you've asked these seven questions across your candidate vendors, the pattern of answers tells you what you need to know:

Low-risk vendors: Answer all seven questions directly, provide written documentation, have security certifications, and offer a private/self-hosted deployment option. Their privacy policy and contract language is specific, not vague.

Medium-risk vendors: Answer most questions but have gaps in data retention policy or subprocessors they can't fully name. They may offer private deployment but haven't thought through the bar ethics documentation. The risk may be manageable with internal firm policy controls.

High-risk vendors: Cannot answer questions 1, 2, or 3 specifically. Point to terms of service as the governing document. Cannot produce security certifications. No private/self-hosted option. These vendors may still be useful for non-confidential internal workflows — but not for anything touching client matters.

What OCI does differently

OpenClawInstall.AI deploys private AI agents for law firms on dedicated servers — your infrastructure, your data, your model choices. Client data does not go to third-party SaaS platforms. Your firm's confidential documents, case files, and client communications stay inside your environment.

Every deployment includes a data handling specification, a private infrastructure architecture document, and a bar ethics alignment framework your firm can use in its own Rule 1.6 analysis.

If your firm is evaluating AI vendors and wants to understand what a genuinely private deployment looks like, request a configuration walkthrough.

Request a Private AI Configuration Walkthrough →  |  Estimate Your ROI →  |  How OCI Compares to Casetext, Harvey, and CoCounsel →