AI Governance and Confidential Data in Law Firms
Introduction : Artificial intelligence has moved quickly from an experimental tool to part of everyday legal work. Lawyers now use AI for research, first drafts, document review, translation, chronologies, and transaction support. But for law firms, the benefits come with a familiar professional question : how do we use a powerful technology without weakening the confidentiality that clients place in us?
The answer is not to prohibit AI altogether. It is to govern it deliberately. A law firm’s duty to protect client information does not change because the information is processed by an AI system. What changes is the risk architecture: the number of places where information can be uploaded, retained, accessed, or used for training.
Why AI changes the confidentiality equation
Traditional legal work already involves sharing confidential information with trusted third parties - counsel, translators, e-discovery vendors, forensic experts, and cloud-service providers. The difference with AI is that information may be processed by a system whose terms of service, data-retention practices, and training arrangements are not always obvious.
A lawyer may paste a clause into a public chatbot, upload a contract for summarisation, or ask an AI tool to review a set of documents. Each action may create a new data trail. Depending on the platform, the input may be retained, reviewed by personnel, disclosed to sub processors, or used to improve the model.
This is why “Do you use AI?” is no longer a sufficient question. Clients increasingly want to know which tools are used, for what purpose, whether their information is uploaded, where it is processed and stored, who can access it, whether it can train a model, and what human review takes place.
Confidentiality and privilege under Indian law
In India, legal professional privilege is primarily governed by Sections 132 to 134 of the Bharatiya Sakshya Adhiniyam, 2023. Section 132 prohibits an advocate from disclosing, without the client’s consent, communications made to the advocate in the course and for the purpose of professional employment. Section 134 gives the client the right to refuse disclosure of such communications.
An AI platform is not an advocate. It is not enrolled under the Advocates Act, 1961, and it does not owe professional duties to the client in the same way as a lawyer. If a client independently uses a public AI tool to analyse a legal issue, there is generally no advocate-client communication and therefore no Section 132 privilege.
The more difficult question arises when a lawyer uses AI while acting for a client. If the AI tool processes an existing privileged communication, the issue is whether the platform’s handling of that information preserves the confidentiality on which privilege depends. Indian courts have not yet directly decided whether uploading privileged material to an AI platform amounts to waiver. The prudent approach is therefore to treat client-identifiable or privileged material as confidential unless the platform provides enforceable safeguards and the client has approved the use.
The Supreme Court’s 2025 decision in In Re: Summoning Advocates who give legal opinion or represent parties during investigation of cases adds another layer. It held that in-house counsel do not receive Section 132 privilege because salaried employment takes them outside the definition of an advocate practising in courts; they retain only the narrower protection under Section 134.
For businesses, this means that AI governance cannot be left only to outside counsel. In-house legal teams must also consider how AI tools handle their own communications, internal legal analyses, and sensitive corporate information.
The IP problem: unpublished inventions and trade secrets
The stakes are particularly high in intellectual-property work. Unpublished patent applications, invention disclosures, technical drawings, source code, manufacturing processes, and proprietary research often have value precisely because they have not been disclosed.
Uploading such material to a public AI tool can create more than a confidentiality issue. If the disclosure is treated as a public disclosure, it may become prior art and affect the patentability of a later-filed application. It may also weaken trade-secret protection if the information is no longer subject to reasonable secrecy measures.
A firm should therefore apply a higher approval threshold before using AI for IP matters. Unpublished patent specifications and invention disclosures should remain out of AI systems unless the client gives written permission and the platform offers enforceable confidentiality, no-training, retention, access-control, and deletion commitments. Trade secrets provided under an NDA should be treated according to the terms of that agreement.
The practical test is simple: the AI environment should be at least as private as the secure systems and physical controls the firm would otherwise use for the same material.
What clients now expect
Clients are no longer satisfied with a general assurance that a law firm “uses secure AI.” They expect transparency about the tools used, the data uploaded, the safeguards in place, and the audit trail available.
Some clients prohibit AI processing altogether for sensitive matters. Others require prior approval, restrict processing to specified jurisdictions, prohibit use of client data for model training, or require disclosure of AI use in work product or invoices. Where a client’s outside-counsel guidelines prohibit AI processing, the firm should follow that instruction even if its own platform is secure.
These requirements should be addressed at the engagement stage and reflected in engagement letters, matter-specific instructions, and vendor contracts. A firm’s internal policy cannot override a client’s stricter requirements.
Building a practical AI governance framework
A useful AI governance framework should make the safe route the easy route. If lawyers must navigate a lengthy policy every time they need help with a research task, many will use whatever tool is available. The firm should therefore provide approved tools, clear rules, and practical training.
A strong framework should include:
Data classification: Identify ordinary business information, confidential client information, privileged communications, personal data, unpublished IP, and trade secrets.
Approved-tool register: Maintain a list of permitted AI tools, permitted use cases, and prohibited inputs for each tool.
Vendor diligence: Review terms on training, retention, human review, sub processors, data location, security, audit rights, breach notification, and deletion.
Data minimisation: Remove client names, identifiers, confidential deal terms, and privileged content where the task does not require them.
Access controls: Limit access by matter team, role, and need-to-know; maintain logs for sensitive matters.
Human verification: Require a lawyer to verify citations, facts, quotations, translations, and legal conclusions before use in advice or filings.
Client instructions: Record client-specific AI restrictions in engagement terms and enforce them through technical controls where possible.
Incident response: Define reporting, containment, notification, and remediation steps for suspected AI-related data exposure.
Training: Train lawyers and staff on approved tools, prohibited inputs, prompt hygiene, verification duties, and escalation.
The DPDP dimension
Where AI tools process personal data, the Digital Personal Data Protection Act, 2023 also becomes relevant. A data fiduciary must take reasonable security safeguards to prevent a personal-data breach, including in respect of processing undertaken by a data processor on its behalf.
The DPDP framework is built around principles such as consent and transparency, purpose limitation, data minimisation, accuracy, storage limitation, security safeguards, and accountability.
For law firms, this means that AI use involving employee data, client contact details, litigation records, or personal information in documents should be assessed through a data-protection lens. The firm should ask whether the processing is necessary, proportionate, secure, and consistent with the purpose for which the information was collected.
The Supreme Court’s draft court regulations
The Supreme Court of India released its draft Regulations for Use of Artificial Intelligence in Courts, 2026 on 3 June 2026. The draft is intended to establish principles of human primacy, transparency, accountability, data protection, and judicial independence in the use of AI across the judicial system.
The draft regulations make clear that AI should assist legal work rather than replace human judgment. They also indicate that sensitive judicial data should not be transferred to external systems without express written authorisation, and that personal data should not be used for training or refining AI systems without appropriate approval.
Although these regulations are directed at courts rather than private law firms, they are a useful signal. They reflect the direction of travel: AI can support legal work, but confidentiality, verification, accountability, and human oversight remain non-negotiable.
Drafting an AI-use policy
A law-firm AI policy should be short enough to be followed and specific enough to be enforced. It should distinguish between permitted, restricted, and prohibited uses.
Permitted uses may include legal research, first-draft summaries, translation support, document organisation, and internal drafting assistance.
Restricted uses should cover client-identifiable material, privileged communications, unpublished IP, deal terms, and regulated personal data. These uses should require an approved enterprise tool and, where appropriate, client consent.
Prohibited uses should include entering sensitive client data into public AI tools, using AI output without verification, and any use that breaches client instructions or professional obligations.
The policy should require lawyers to record the tool used, the data category, the purpose, and any client approval for sensitive matters. It should also state clearly that AI output is a draft or research aid, not a substitute for legal analysis.
What businesses should ask outside counsel
Businesses should ask practical questions before engaging a law firm on sensitive matters:
Which AI tools are approved for this matter?
Will client data be uploaded, and if so, what data will be removed first?
Where will the data be processed and stored?
Can the vendor use the data to train or improve models?
Who can access the data, including sub processors?
What retention, deletion, audit, and breach-notification terms apply?
What human review will occur before advice or work product is delivered?
Does the firm need written client consent for the proposed use?
These questions help businesses assess whether the firm’s AI use is consistent with confidentiality, privilege, trade-secret protection, and data-protection obligations.
The bottom line
AI can improve legal productivity, but it should be governed like any other third-party processor of confidential information. A law firm’s responsibility remains unchanged: protect client confidences, preserve privilege where applicable, verify AI-assisted work, and follow client-specific restrictions.
The safest approach is not a blanket ban on AI, but a controlled framework: approved tools, data classification, contractual safeguards, access limits, human review, training, and clear client communication. That approach allows firms to capture efficiency without treating client trust as an acceptable trade-off.
Author: Vipasha Srivastava in case of any queries please contact/write back to us via email to content@khuranaandkhurana.com or at Khurana & Khurana, Advocates and IP Attorney
Endnotes
Section 132, Bharatiya Sakshya Adhiniyam, 2023, protects confidential communications made to an advocate in the course and for the purpose of professional employment.
Section 134, Bharatiya Sakshya Adhiniyam, 2023, confers on the client the right to refuse disclosure of confidential communications with a legal professional adviser.
In Re: Summoning Advocates who give legal opinion or represent parties during investigation of cases, 2025 INSC 1275, holding that in-house counsel do not receive Section 132 privilege and retain only the narrower protection under Section 134.
Digital Personal Data Protection Act, 2023, Section 8(5), requiring a data fiduciary to take reasonable security safeguards to prevent a personal-data breach, including processing undertaken by a data processor.
Press Information Bureau, “DPDP Rules, 2025 Notified,” noting the Act’s principles of consent and transparency, purpose limitation, data minimisation, accuracy, storage limitation, security safeguards, and accountability.
Supreme Court of India, Notice dated 3 June 2026, releasing the draft Regulations for Use of Artificial Intelligence in Courts, 2026 for stakeholder comments; the deadline was subsequently extended to 15 July 2026.
Supreme Court of India, draft Regulations for Use of Artificial Intelligence in Courts, 2026, reflecting principles of human primacy, transparency, accountability, data protection, and judicial independence.
ET Legal World, “Law Firms Tighten Data Controls as AI Adoption Surges, With IP Work Facing High Stakes,” 15 September 2026, discussing law-firm practices on approved AI environments, data retention, model-training restrictions, access controls, and client AI policies.
AZB & Partners, “AI And Its Impact On Legal Privilege,” 15 July 2026, analysing the interaction between AI, confidentiality, and legal professional privilege under the Bharatiya Sakshya Adhiniyam, 2023.
Comments