Copilot for Microsoft 365, Copilot Cowork, and Claude Cowork: What's the Difference — and Why Getting It Wrong Could Cost Your Firm Attorney-Client Privilege
A federal judge just ruled that a criminal defendant lost attorney-client privilege because he used a consumer AI tool with client information. Here's what law firms need to know about the three AI tools causing the most confusion right now.
In United States v. Heppner, Judge Jed Rakoff (S.D.N.Y.) ruled that 31 documents created using a consumer version of Claude were not protected by attorney-client privilege. The moment the defendant typed client information into a public AI platform, he disclosed it to a third party. The government got the documents.
Why the AI Tool You Choose Is Now a Legal Decision, Not Just an IT One
Before the tools: the Heppner ruling matters for every law firm in America, and it's worth understanding precisely what the court held — and what it didn't.
Judge Rakoff ruled that Bradley Heppner's AI-generated documents failed to qualify for attorney-client privilege on three grounds. First, the communications were not between a client and an attorney — Claude is not a lawyer. Second, they were not confidential — Heppner used a consumer version of Claude whose privacy policy permitted Anthropic to use inputs and outputs for model training and to disclose data to third parties, including government authorities. Third, he did not use the AI tool at the direction of counsel.
The court explicitly noted the outcome might have been different if counsel had directed Heppner to use Claude — which could have established the AI as the lawyer's agent under the Kovel doctrine. Enterprise AI tools with contractual confidentiality provisions were also not addressed. This is a warning about consumer AI misuse, not a prohibition on AI in legal work.
ABA Model Rule 1.6 has required attorneys to take reasonable measures to prevent unauthorized disclosure of client information for years. Heppner makes clear what "reasonable measures" must now include: understanding the privacy terms of every AI tool your team touches, and keeping client data inside governed, enterprise environments where confidentiality obligations are contractually enforceable.
What We're Already Seeing at Law Firms
Here's what a typical week looks like for a small law firm in Silicon Valley in 2026: an associate summarizes deposition notes using the free version of ChatGPT on a personal laptop. A paralegal uses Claude's consumer plan to draft a response letter. A partner installs an AI writing app without telling anyone. None of them intended to create a problem. None of them realized they just disclosed client information to a third-party AI provider operating under consumer terms of service.
Consumer-tier AI plans — free, Pro, or personal subscriptions — share characteristics that make them inappropriate for client work:
- Terms typically permit the provider to use your inputs for model training
- No Data Processing Agreement (DPA) and no contractual confidentiality provisions
- Unlike Microsoft 365 or Google Workspace, the provider is not contractually bound to act only as a processor of your data
- Consumer terms often grant broad rights to process, retain, and disclose what you share
Some discussions have described Claude's Computer Use feature as capturing continuous video of your screen at several frames per second. That's inaccurate. Computer Use is currently available only to commercial API customers — not consumer plan users — and captures targeted screenshots to interpret specific tasks, not a continuous stream. The risk with consumer AI is real; it doesn't need to be overstated to be worth taking seriously.
The Three Tools — What Each One Actually Is
Copilot for Microsoft 365
AI assistant built into Outlook, Word, Teams, Excel, and PowerPoint. Responds to one prompt at a time inside your existing apps.
Copilot Cowork
Multi-step task execution across all M365 apps. Give it a goal; it builds a plan and runs it in the background with your approval.
Claude Cowork
Multi-step task execution across any app on your computer. Runs locally on your device — outside your Microsoft 365 governance environment.
Copilot for Microsoft 365 — the details
Copilot for M365 is conversational and single-task: you prompt it, it responds. In a law firm it helps summarize email threads, draft client letters from bullet points, generate Teams meeting recaps with action items, and surface document insights — all inside your existing M365 environment.
Is it safe for client work? When properly configured inside your Microsoft 365 tenant, yes. Your prompts and outputs are not used to train Microsoft's AI. Client data stays inside your firm's environment. Copilot only sees what the user already has permission to access. Microsoft's enterprise Data Processing Addendum confirms these commitments contractually — which is exactly the kind of enforceable protection the Heppner analysis turns on.
These protections apply to Microsoft 365 Copilot — the business or enterprise version, not to the free consumer Copilot chat experience at copilot.microsoft.com. And "properly configured" requires that transcription settings, data retention policies, and user access controls have been reviewed by someone who understands how they interact.
Copilot Cowork — Next-Generation AI Execution
Announced March 9, 2026 as part of Microsoft's Wave 3 release, Copilot Cowork moves from AI assistance to AI execution. Instead of asking Copilot to draft a meeting agenda, you tell Cowork: "Prepare for the Henderson client review on Thursday." It reviews the client's recent emails, pulls documents from SharePoint, drafts the agenda, summarizes open items from the last meeting, and blocks prep time on your calendar — all from one instruction, running while you focus on other work.
Built in close collaboration with Anthropic, Copilot Cowork runs Claude's AI model inside Microsoft's cloud infrastructure. All processing happens within your firm's Microsoft 365 tenant — covered by Microsoft's enterprise data protections, governance policies, and compliance framework. Nothing is processed on a personal device or through a consumer account. For law firms with client confidentiality obligations under ABA Rule 1.6, that architectural distinction matters enormously.
Claude Cowork — Powerful, But Read the Fine Print
Claude Cowork is Anthropic's standalone product — separate from anything Microsoft makes. Released for Mac in January 2026 and Windows in February 2026, it is designed for non-technical workers who want to delegate complex, multi-step tasks to an AI that can see and interact with files and applications open on their screen across any app, not just Microsoft 365 ones.
It is genuinely capable. But for law firms, where it operates matters as much as what it can do. Claude Cowork runs locally on your device, outside your firm's Microsoft 365 governance environment. The risk depends entirely on which Claude plan your firm is using:
- Consumer plans (Pro, free) — No DPA, broad data rights in the terms of service. Inappropriate for client-related work. This is what Heppner addressed.
- Claude for Business / Enterprise plans — Include contractual confidentiality and no-training provisions. Materially different risk profile, but still requires attorney direction and a written AI use policy.
Side-by-Side: Which Tool Is Which
| Copilot for M365 | Copilot Cowork | Claude Cowork | |
|---|---|---|---|
| Made by | Microsoft | Microsoft + Anthropic | Anthropic |
| What it does | AI assistant in M365 apps | Multi-step task execution in M365 | Multi-step across all apps |
| Runs where | Microsoft cloud | Microsoft cloud | Your local device |
| Works with | Outlook, Word, Teams, Excel | All M365 apps | Any app on your computer |
| Enterprise DPA | ✔ Yes | ✔ Yes | ⚠ Business/Enterprise only |
| Data used for training | ✔ No (enterprise) | ✔ No (enterprise) | ⚠ No on Business; yes on consumer |
| Privilege risk | Low ✔ | Low ✔ | High on consumer ✖ |
| Availability | Now | Rolling out Mar–Apr 2026 | Now (Mac + Windows) |
| Price | $21/user/mo add-on | Included with Copilot | From $20/user/mo (separate) |
What Should Your Firm Do Right Now
Most small law firms don't need to evaluate all three tools simultaneously. Here's the practical sequence:
Audit what your team is already using
Find out which AI tools are on personal devices, work laptops, and browser extensions. You can't govern what you don't know exists. Shadow AI is the most common privilege risk at small firms right now.
Confirm your Microsoft 365 Copilot status
Find out if your firm has the Copilot add-on active. If it's included in your plan and hasn't been configured, you're leaving it on the table — and possibly leaving your team to fill the gap with consumer tools.
Create a written AI acceptable use policy
Which tools are approved for client-related work? Which require IT review first? What happens when someone is unsure? A brief, clear policy reviewed by your managing partner and IT provider prevents the majority of privilege risks.
Watch Copilot Cowork as it rolls out
When Cowork becomes broadly available over the next few months, it will meaningfully change how firm administrators handle complex, multi-step coordination tasks — inside a governed environment your firm already controls.
Not sure where your firm stands?
eSudo has helped small law firms in Silicon Valley and San Francisco navigate IT security and compliance since 2001. We'll review your Microsoft 365 setup, confirm your AI governance posture, and give you a clear picture of where your risks are — usually in a single conversation.
Frequently Asked Questions
This article references an evolving area of law and rapidly changing AI product features. Pricing, availability, and legal interpretations reflect information available as of March 2026. Verify current Microsoft and Anthropic pricing before making purchasing decisions. This article is for informational purposes only and does not constitute legal advice. Schedule a review date for September 2026.