The AI never holds the identifiers, and documents are screened for MNPI before they reach it. One tool prepares court productions to Rule 5.2. The other lets your people use AI on real documents while names and numbers stay in your own cloud.
Last updated
This page covers three things. A plugin for Claude prepares court filings, redacting to Federal Rule of Civil Procedure 5.2. A privacy gateway swaps names and numbers for placeholders before text reaches an AI model. MNPI screening inside the gateway passes, holds for review or blocks each document. In all three, a person decides what matters.
Client identifiers such as names, Social Security numbers and account numbers end up in AI tools and in court filings. So does material non-public information. It happens because people paste and file what is in front of them.
Identifiers and MNPI, material non-public information, are different problems. One is who a document is about. The other is what it reveals about a deal or a position. So the gateway handles them in two steps: it swaps the identifiers, then screens the text. Read how a private AI workspace works or How to keep MNPI out of public AI tools.
The Court Case File plugin for Claude prepares case documents for filing or production. You point it at a folder and say "prepare these for court." It runs inside Claude on the attorney's own computer, shows its full plan first, and writes nothing until told to go. It is offered today.
Originals are read where they sit, processed in memory, and never uploaded or stored anywhere new.
Identifiers are removed for good, following Federal Rule of Civil Procedure 5.2. This describes the rule the plugin follows. It is not legal advice.
Only the last four digits remain.
Only the year remains. Other dates are left alone.
Reduced to initials when the text clearly marks a minor. Anything uncertain is flagged for the attorney.
Bank, card and routing numbers keep only the last four digits.
The rule is a floor. The default policy also fully redacts addresses, emails, phone numbers, license and passport numbers and IP addresses. A minimal preset applies exactly the rule's four requirements. The firm chooses per case.
Every page, in sequence.
One row per document, with no identifier value in it.
Documents that look privileged are flagged and queued for a person.
The reviewing attorney personally certifies every production, including every flagged and withheld document.
The tool never decides what is privileged. It surfaces candidates and makes no legal determination.
A practice run shows what it found, what it would redact and what it set aside. Nothing is written until approved.
The privacy gateway sits between your firm and the AI model, for everyday AI use. Before text reaches the model, names, Social Security numbers, account and routing numbers, emails and phone numbers are swapped for placeholders. The model works on the substance. It removes who and keeps what, so amounts, dates and substance pass through and the analysis stays useful.
It is deployed per firm, into the firm's own Azure environment, as part of an engagement.
Inside your firm's Azure environment
Outside
The AI model receives placeholders, and the real values stay in your environment. DoubleChecked receives only anonymous usage counts.
It works on PDF, Word, Excel and scanned documents, not only pasted text.
Staff still decide what to ask and what to rely on. They check the answer before it goes anywhere. The gateway does not decide what a document means, and a person reviews anything the MNPI screening holds.
MNPI screening is a step inside the gateway. It runs after identifiers are swapped for placeholders, inside the firm's own Azure environment, on the firm's own model deployment. Nothing leaves the firm's environment for this check.
The document goes on to the AI model.
A person can release it after confirming.
The document is quarantined. The gate cannot release it.
The firm decides what counts as MNPI for it.
The policy is set for each firm at deployment.
When the screening is unsure, it holds the document for review.
No setting makes it pass documents through when the check cannot run.
These are illustrations of where the two tools could be used, grouped by industry. They are examples of what is possible, not descriptions of client work.
Bring a sample. We will show what it finds and what it leaves.
Every project gets a fixed price after one working session, and you know the number before we build.
The AI model receives placeholders, and the real values stay in your firm's environment. For court filings, document text goes only to the firm's own Azure AI service in the firm's own subscription, to find identifiers. DoubleChecked never sees it.
Originals are read where they sit, processed in memory, and never uploaded or stored anywhere new. The redacted production is the one thing meant to leave. For the gateway, documents and the placeholder vault stay in your Azure environment.
Yes, inside the gateway. It makes three passes: language, structure and a model adjudicator. Each document passes, is held for a person, or is blocked. The firm defines what counts as MNPI for it.
It holds the document for human review. A person can release it after confirming. A block cannot be released by the gate. It fails closed.
A person does. The plugin flags documents that look privileged and queues them for review, and it drafts the log. It never decides what is privileged. The reviewing attorney certifies every production, including every flagged and withheld document.
Social Security and taxpayer numbers keep the last four digits. Birth dates keep the year. A minor's name becomes initials. Financial account numbers keep the last four digits. This describes the rule the plugin follows, not legal advice.
Claude, and your firm's own Azure subscription. The plugin runs inside Claude on the attorney's computer. The gateway is deployed into your Azure environment as part of an engagement, with sign-in and roles set up for your people.
Every project gets a fixed price after one working session, and you know the number before the build starts. A working session is 45 minutes on one idea from your firm. We sign an NDA before it.
Your own AI environment, where company data stays inside.
Documents read, sorted and checked.
How we handle data, access and review.
The rest of what we offer.
An insight on keeping MNPI out of AI tools.
A working session is 45 minutes on one idea from your company: what to build, what it needs, and what it would cost.
