Key takeaways
- Built-in DLP patterns detect common identifiers such as Social Security numbers, not a firm's internal client and matter numbers.
- A Microsoft Purview custom sensitive information type combines a primary pattern with supporting keywords within a set character proximity and assigns confidence levels.
- Exact data match compares content against a hashed table of the firm's real records, which Microsoft says can hold up to 100 million rows and be refreshed daily.
- Microsoft documents that DLP can stop Microsoft 365 Copilot from responding to prompts that contain chosen sensitive information types, a feature it lists as in preview.
- California's confidentiality duty and the technology competence comment to rule 1.1 give managing partners reason to make these controls reliable.
Why should a DLP policy know your client and matter numbers?
Client and matter numbers mark the documents a law firm most needs to protect, and standard data loss prevention rules do not recognize them. Out of the box, DLP looks for Social Security numbers, card numbers and similar identifiers. A pleading, a settlement memo or a privileged email often contains none of those, yet nearly every one carries a client and matter number in the file name, the footer, the subject line or the document management profile.
That makes the matter number a dependable signal. If DLP can recognize it, the firm can warn an attorney before matter content goes to a personal email account, stop a file upload to a public AI tool, and show who tried. Without it, the firm is relying on policy and memory.
What do the confidentiality rules expect of the firm?
California's confidentiality rules set the duty and leave the choice of tools to the firm. Business and Professions Code section 6068(e)(1) requires a California attorney to maintain inviolate the confidence, and at every peril to preserve the secrets, of the client. Comment [1] to rule 1.1 ties competence to keeping abreast of the benefits and risks of relevant technology. Rules 5.1 and 5.3 require lawyers with managerial authority to make reasonable efforts to ensure the firm has measures giving reasonable assurance that lawyers and nonlawyers comply.
For AI specifically, ABA Formal Opinion 512 (persuasive in California, not binding) says self-learning generative AI tools raise the risk that one client's information is disclosed improperly, and that managerial lawyers should set clear policies on permitted use and train lawyers and staff. A DLP rule that recognizes matter numbers is one way to make that policy enforceable. None of these sources requires a particular product, and whether a control is reasonable for your firm is a judgment for its own ethics counsel.
How do you build a rule for your matter number format?
Write the format down, then turn it into a custom sensitive information type in Microsoft Purview. Microsoft's documentation describes each pattern as a primary element, such as a regular expression or keyword list, plus optional supporting elements found within a character proximity you set, with low, medium and high confidence levels. Work through these steps:
- Pull the real format from your practice management system: digits, separators, prefixes, and any legacy formats from before a system change or a merger.
- Write a pattern for each format as the primary element. Microsoft's guidance says not to use the start and end anchors common in other regex tools, and to keep a single capturing group.
- Add supporting keywords that appear near real matter numbers in your documents, such as "matter," "client no.," "file no." or your firm's name.
- Set confidence: a pattern alone at low, the pattern with a keyword nearby at high.
- Test against a sample folder of real documents and a sample of documents that should not match.
How do you keep false positives from burying the real alerts?
Require corroborating evidence and, where you can, match against real records. A bare number pattern collides with other numbers. A five-digit number, a hyphen and four more digits also describes a ZIP+4 code, and many invoice, check and docket numbers look similar. Rules that fire on every address block get ignored within a week.
Keyword proximity helps. Exact data match helps more. With exact data match, the firm uploads a table of its actual client and matter records, which Purview hashes so the values are not stored in readable form. Microsoft says the table can hold up to 100 million rows and be refreshed daily. A number that only looks like a matter number will not match, and the rule can require a client name nearby as corroboration. Microsoft recommends using the exact data match type at higher confidence and the pattern-based type at lower confidence, so you catch both exact hits and broader occurrences.
Where should a law firm's DLP policy apply?
Apply it everywhere matter content can leave the firm. That usually means Exchange email, SharePoint and OneDrive sharing, Teams chat, firm devices through endpoint DLP, and browsers, so uploads and pastes to unapproved AI sites are covered. Microsoft also documents a DLP location for Microsoft 365 Copilot that can stop Copilot from responding to prompts that contain chosen sensitive information types, a feature Microsoft lists as in preview, and can block web searches for those prompts. Microsoft notes that this check reads the typed prompt and does not scan files uploaded directly into it, so the endpoint and browser rules still matter.
Should the DLP policy warn or block?
Start by watching, then warn, then block where the risk is clear. Run the policy in audit mode for two to four weeks and review what it catches. Turn on policy tips that explain the rule and let attorneys override with a business reason for internal sharing and approved vendors. Block outright for uploads to public or unapproved AI tools and personal accounts, where there is rarely a good reason for matter content to go.
What should you do this quarter?
- Document every client and matter number format in use, including legacy ones.
- Build and test a custom sensitive information type with keyword support.
- Decide whether exact data match is worth it, and who owns the export from practice management and its refresh.
- Run the policy in audit mode across email, files, Teams, devices and browsers, then set warn and block actions.
- Add the rule to your AI policy and attorney training so the warnings make sense when they appear.
- Review overrides monthly and tune the rule.
How does s90 help?
s90 builds and tunes DLP for law firms: mapping your matter numbering, writing and testing the detection rules, extending them to email, files, devices, browsers and Copilot, and reviewing what they catch so the alerts stay useful. If you want a rule set your attorneys will not learn to ignore, we are glad to help.
