Here’s the short answer. Your documents stay in NetDocuments. The text of any document Claude reads does go to Claude.
Both of those things are true, and you need both to answer your firm’s security question. NetDocuments’ marketing says “documents never leave the platform,” and that’s accurate as far as it goes. But if Claude is going to summarize an engagement letter, Claude has to read it. Here’s what actually happens.
What Happens When Claude Reads a Document
When you connect Claude to NetDocuments through the MCP connector, nothing gets copied over in bulk. NetDocuments doesn’t sync your repository to Anthropic, and Claude doesn’t keep a copy of your cabinets.
When you ask Claude a question, this is what happens:
- Claude searches NetDocuments as you, using your own NetDocuments login.
- NetDocuments checks your permissions, ethical walls, and DLP rules before it returns anything.
- NetDocuments sends the relevant content to Claude as text, and only the documents you already have access to.
- Claude uses that text to answer you, and the search is recorded in the NetDocuments Consolidated Activity Log.
The original files never move. They stay in NetDocuments, with your version history, your profile attributes, and your security. But the text Claude reads is now part of your Claude conversation.
NetDocuments says this directly on its ndConnect page: what the AI vendor does with the content after delivery is governed by your organization’s agreement with that vendor. So the real question isn’t whether data leaves NetDocuments. It’s what Anthropic does with it once it’s in Claude.
What Claude Does With It on a Team Plan
Claude Team and Enterprise run under Anthropic’s Commercial Terms, not the consumer terms that cover Pro and Max. That’s why I tell firms to connect Claude to NetDocuments on a Team or Enterprise account, not someone’s personal plan.
On a Team plan:
- Your data isn’t used to train Claude. Anthropic doesn’t train its models on inputs or outputs from its commercial products unless your firm opts in to its Development Partner Program. (Anthropic)
- Your firm is the data controller. Under the Commercial Terms, the firm controls the data its users submit, and Anthropic processes it on the firm’s behalf. (Anthropic)
- Conversations stay until you delete them. Chats are kept in Claude so users can go back to them. When someone deletes a chat, it disappears from their history right away and is deleted from Anthropic’s back end within 30 days. (Anthropic)
- Incognito chats are deleted automatically within 30 days.
- Flagged chats are kept longer. If Anthropic’s trust and safety systems flag a chat as a Usage Policy violation, its inputs and outputs can be kept for up to two years.
Turn Off Chat Ratings
When a user clicks thumbs up or thumbs down on a response, that conversation is sent to Anthropic as feedback and kept for five years. For a law firm, I’d turn that off.
An Owner or Primary Owner can do this in Organization settings → Data, under the Rate chats setting.
Team vs. Enterprise
On a Team plan, the firm doesn’t set a retention period. Chats stay until a user deletes them. Claude Enterprise lets an Owner set a custom retention period (30 days minimum) so chats and projects are deleted automatically. If your firm’s retention policy requires that, it’s a reason to look at Enterprise.
Has Anyone’s Client Data Actually Leaked?
I haven’t found a reported case of a law firm on a Claude Team or Enterprise plan having client data leak out of Claude. Here’s what has been reported in 2026, and what each one means for your firm.
The Heppner Ruling: Consumer Claude Isn’t Confidential
In February 2026, Judge Rakoff in the Southern District of New York ruled that a criminal defendant’s conversations with Claude weren’t protected by attorney-client privilege or the work product doctrine. Part of the reasoning was that the consumer privacy policy he agreed to allowed Anthropic to collect and disclose his inputs, so he had no reasonable expectation of confidentiality. (National Law Review)
That was a client using a consumer account on his own, not a lawyer on a firm account. Courts haven’t said whether a Team or Enterprise account would come out differently. But this is exactly why your firm shouldn’t be doing this on Pro or Max.
Shared Chats Showed Up on Google
In July 2026, Claude conversations and Artifacts showed up in Google search results. These were chats people had shared with the Share button, which creates a public link. Private chats weren’t exposed. (TechCrunch)
The fix is simple: don’t create public share links for chats with client content in them. You can review and revoke shared links in Settings → Privacy → Shared chats.
Anthropic’s Account Data Was Accessed
In October 2026, an outside security firm got into some of Anthropic’s internal business systems. Anthropic notified affected customers that business contact names, email addresses, account and organization IDs, plan tier, and usage and spend figures may have been exposed. Anthropic says the content of conversations with Claude wasn’t included. (The Elec)
That’s a real incident, but it was billing and account information, not anything your attorneys typed into Claude or anything Claude read from NetDocuments.
What This Doesn’t Cover
- What users paste or upload. The connector respects NetDocuments permissions. It doesn’t control what someone copies from another source into a chat.
- Where the answer goes next. If an attorney takes a draft Claude wrote and saves it on their desktop, that’s a document management problem, not a Claude problem. Save it back into NetDocuments.
- Your ethics obligations. Whether your firm needs client consent to use AI on a matter is a question for your firm and your bar, not for me.
Sources
NetDocuments
Anthropic
- Is my data used for model training?
- How long do you store my organization’s data?
- Does Anthropic act as a data processor or controller?
- Configure custom data retention controls for Enterprise plans
The Heppner ruling
- National Law Review – “Claude Is Not an Attorney”
- Duane Morris – The Perils of Privilege Waivers Through AI
Shared chats on Google
- TechCrunch – Your Claude shared chats and Artifacts may have ended up on Google
- 404 Media – Tons of People’s Claude Chats and Creations Are Exposed on Google
Anthropic account data incident
- The Elec – Anthropic Faces Series of Enterprise Data and Developer Credential Leaks
- MLQ – Anthropic emails enterprise customers that an outside security firm may have viewed their contact, plan and spend data
Need Help?
Schedule a free consultation at go.oncehub.com/optconsult, or reach us at optiable.com/contact.
About the Author
Craig Bayer is the founder and leader of Optiable, an award-winning document management (DMS) consulting firm dedicated to helping law firms seamlessly integrate NetDocuments. Specializing in firms with 10 to 150 users, he has rolled out NetDocuments for over 600 law firms across the United States and Canada since 2010.
With deep expertise in the legal industry, Craig has a proven track record of optimizing technology to meet the unique needs of law firms. His certifications include industry-leading tools such as Amicus Attorney, Centerbase, Clio, PCLaw, HotDocs, TimeMatters, Soluno, and Worldox, enabling him to deliver comprehensive solutions tailored to each client’s workflow and goals.
CONNECT WITH CRAIG → https://www.linkedin.com/in/craigbayer/

