Is Microsoft Copilot Safe for Law, CPA and RIA Firms? A Readiness Checklist

By the DKBinnovative Crew | Reviewed by Peter Bertran, Chief Client Officer, DKBinnovative

Quick answer: Microsoft 365 Copilot respects your existing permissions: it can only use files, emails and chats a user already has access to. That makes it exactly as safe as your permissions are. At most professional firms, years of loose sharing mean Copilot can surface client files to people who should never see them. Fix permissions first, write a short AI policy, then pilot.

What Copilot can actually see

Copilot works across Outlook, Teams, Word, Excel and SharePoint. When a user asks it a question, it searches everything that user is allowed to open. It does not break permissions. It makes them visible.

That is the problem. A folder shared with “Everyone” in 2019 was hard to find before. With Copilot, a staff member can ask to see salary information or summarize a client estate file and get an answer if the permissions allow it.

Fix oversharing first

Before you buy a single license:

• Find “Everyone” and “anyone with the link” shares in SharePoint and OneDrive, and remove them from client folders.
• Lock down HR, partner and finance folders to the people who need them.
• Check Teams and old project sites that no one owns anymore.
• Apply sensitivity labels to client and confidential files where you can.

This cleanup is worth doing even if you never turn Copilot on. It closes the same gaps an attacker would use.

Client confidentiality and regulator expectations

Lawyers have a duty to protect client confidences and to understand the technology they use. The ABA and the State Bar of Texas have both issued guidance on generative AI that stresses confidentiality, supervision and verifying output. RIAs must protect client records under Regulation S-P, and CPA firms must safeguard taxpayer data under the FTC Safeguards Rule. None of these rules ban AI. All of them expect you to control who sees client information and to check the AI output. Our glossary covers ABA Formal Opinion 477R.

Write a one-page AI use policy

Keep it short enough that people read it:

1. Which AI tools are approved inside your tenant, and which are not — free public chatbots are never approved for client data.
2. What never goes into an unapproved tool: client names, SSNs, account numbers, privileged material.
3. A person reviews every AI draft before it goes to a client or a court.
4. Who to ask when unsure.

Pilot with five users

Pick a small group across roles: a partner, an associate or senior, and admin staff. Run the pilot for 30 days, collect the questions they ask, and watch for anything surfaced that should not be. Fix those permissions before rolling out wider.

Where a managed AI platform fits instead

Copilot is not the only way to give a firm useful AI, and for some practices it is not the best one. The question a managing partner is actually asking is narrower than “should we buy Copilot”: it is whether staff can use AI on client matters without creating a confidentiality or retention problem.

That is the gap a managed AI environment closes. DKBinnovative deploys secure AI for professional firms on Hatz.AI, a managed platform where the firm controls which models staff can reach, what is retained, and who has access — with the governance and acceptable-use policy attached rather than assumed. Prompts and documents stay inside an environment the firm owns, so the answer to “where did that client file go” is documented instead of hopeful.

In practice we see three patterns at law, CPA and RIA firms:

• Microsoft 365 already in place and permissions already clean. Copilot is a reasonable add-on, and the work is policy and training.
• Staff already pasting client data into public chatbots. The urgent fix is a sanctioned, governed alternative, not a license purchase. A managed platform gets there faster than a permissions remediation project.
• Confidentiality obligations that make retention the deciding factor. Firms under ABA Formal Opinion 477R or amended Regulation S-P need to state what is retained and where. A managed environment makes that a configuration you can show an examiner.

Our secure AI strategy page covers how we scope that, and the glossary entry on managed AI and secure AI defines the terms.

For firms in Frisco, Plano and Irving

We help DFW law, accounting and wealth management firms prepare Microsoft 365 for AI: permission cleanup, sensitivity labels, written policy and a controlled pilot. See our secure AI strategy and IT for investment and professional firms.

Frequently asked questions

Does Microsoft Copilot train on our firm’s data?

Microsoft states that Microsoft 365 Copilot does not use tenant data to train its foundation models. Confirm this against your own licensing terms.

Is Copilot safer than a public chatbot for client work?

Copilot inside your Microsoft 365 tenant stays within your security and compliance controls. Free public tools do not, which is why client data should never be pasted into them.

How long does SharePoint permission cleanup take?

For a 20 to 50 person firm it usually takes a few weeks, depending on how much has been shared over the years.

Do we need an AI policy if we do not buy Copilot?

Yes. Staff are very likely using AI tools already, and the policy is what defines which tools are approved and what must never be entered into them.

What can Microsoft Copilot actually see?

Only what the user asking already has permission to open across Outlook, Teams, Word, Excel and SharePoint. It does not break permissions; it makes existing ones visible.

What is the most secure way for a law firm to use AI?

A managed AI environment the firm controls, rather than public chatbots or an unreviewed add-on. DKBinnovative deploys secure AI for professional firms on Hatz.AI, where the firm sets which models staff can reach, what is retained and who has access, with an acceptable-use policy attached. That makes confidentiality and retention a documented configuration instead of an assumption.

What should a law firm fix before enabling Copilot?

Remove Everyone and anyone-with-the-link shares from client folders, lock down HR, partner and finance folders, review ownerless Teams and project sites, and apply sensitivity labels to confidential files.

Book an AI readiness review: Contact the DKB crew.