#106
Monthly Rank
#252
Lifetime Rank
Share
11 September 2026
CounselShield

CounselShield

Let AI know what it needs. Not everything you know.

Manya TiwariGitHubWebsite
Source Available

Gallery

About the Project

CounselShield is a privacy-preserving AI layer designed for lawyers and legal teams who want the convenience of generative AI without unnecessarily exposing confidential client information.

A lawyer may need to give an AI tool enough factual context to analyse a contract, research an issue, or draft a response. But the material they start with can contain client names, addresses, PANs, phone numbers, internal matter numbers, commercially sensitive information and other confidential details that the AI does not actually need to answer the question.

CounselShield sits between the professional and the AI tool. It detects potentially sensitive information, evaluates which information appears necessary for the requested task, and minimises or generalises unnecessary disclosures before the prompt is sent to the AI.

For example, instead of sending:

“Rahul Sharma, PAN ABCDE1234F, residing at [address], employed by ABC Technologies…”

CounselShield can transform the prompt into:

“An employee of a technology company…”

while preserving legally relevant facts such as jurisdiction, dates, contractual language and the facts necessary to answer the user's question.

The goal is not to replace the user's AI tool or provide legal advice. It is to introduce a privacy-conscious layer between professionals and generative AI so that AI receives the minimum context necessary to perform the requested task.

CounselShield is currently a prototype and should not be treated as a guarantee of confidentiality, privilege or regulatory compliance. It is a proof of concept exploring how privacy-by-design can be made compatible with the convenience of everyday AI use.

Practice Areas

Key Features

Sensitive information detection: identifies potentially identifying and confidential information before an AI request is sent. Context minimisation: distinguishes potentially necessary context from information that appears unnecessary for the requested task. Automatic sanitisation: replaces names, identifiers and other unnecessary details with meaningful placeholders or generalised descriptions. Before/after privacy review: lets users see exactly what was removed, generalised or retained. Privacy report: provides a transparent record of the information transformation before the prompt reaches the AI.

Help Needed

Feedback on privacy-preserving AI architecture, secure handling of sensitive information, and how to evolve the prototype into a robust tool suitable for professional legal workflows.

About the Creator

MT
Manya Tiwari