Legal Confidentiality in the Age of AI - Can Lawyers Safely

A Legal and Ethical Guide for Advocates in India

Artificial Intelligence is rapidly becoming part of professional legal practice.

Lawyers are using AI tools for legal research, drafting, summarisation, document review, proofreading, translation, case-law analysis, contract review and administrative work. These tools can significantly improve efficiency.

But legal practice involves something that ordinary business activity often does not: highly confidential and privileged information.

A lawyer may have access to:

• confidential client communications;
• litigation strategy;
• draft pleadings;
• privileged legal advice;
• commercial agreements;
• financial information;
• personal data;
• medical records;
• investigation material;
• trade secrets;
• settlement negotiations; and
• documents relating to ongoing criminal or regulatory proceedings.

This raises an increasingly important question:

Can lawyers use AI tools without compromising client confidentiality and professional obligations?

The answer under Indian law is not that AI tools are prohibited. There is currently no general Indian rule that says an advocate cannot use artificial intelligence for legal work.

However, the use of AI does not reduce or transfer the advocate's existing professional obligations. If confidential information is entered into an AI system, the advocate remains responsible for ensuring that the use of that system does not result in an unauthorised disclosure or misuse of client information.

1. The Fundamental Principle: AI Does Not Change the Lawyer's Duty

The starting point is simple:

A lawyer's duty of confidentiality exists regardless of the technology used to perform the work.

The Bar Council of India Rules prescribe standards of professional conduct and etiquette for advocates. Rule 17 states that an advocate shall not, directly or indirectly, breach the obligations imposed by Section 126 of the Indian Evidence Act.

The statutory framework has now moved from the Indian Evidence Act, 1872 to the Bharatiya Sakshya Adhiniyam, 2023 (BSA), which came into force on 1 July 2024.

Section 132 of the BSA deals with professional communications.

It provides that, subject to specified exceptions, an advocate cannot disclose a communication made to the advocate by or on behalf of the client in the course and for the purpose of professional service, disclose the contents or condition of documents learned in that professional capacity, or disclose legal advice given to the client, without the client's express consent. The obligation continues even after the professional service has ended.

This is highly relevant to AI.

If a lawyer gives confidential client material to an external AI service, the issue is not simply:

"Did the lawyer personally disclose the information?"

The more important question is whether the lawyer has allowed confidential information to be processed or disclosed through a third-party system in a manner inconsistent with the lawyer's legal and professional obligations.

---

2. What Does Section 132 of the Bharatiya Sakshya Adhiniyam Protect?

Section 132(1) of the BSA protects:

1. communications made by or on behalf of the client;
2. documents with which the advocate becomes acquainted in the course of professional service; and
3. advice given by the advocate to the client.

The protection is subject to important exceptions, including communications made in furtherance of an illegal purpose and facts observed by the advocate during professional service showing that a crime or fraud has been committed since the commencement of the service. The obligation continues after the professional engagement ends.

This is not merely a matter of professional courtesy.

It is part of the legal framework governing the advocate-client relationship.

---

3. The Bar Council of India Rules Also Matter

The BCI Rules independently impose professional obligations upon advocates.

Rule 17 specifically prohibits an advocate from directly or indirectly breaching the obligations relating to professional communications.

The Rules also state that an advocate should not do anything whereby the advocate abuses or takes advantage of the confidence reposed in the advocate by the client.

The Delhi High Court has also recognised the confidential and privileged nature of documents generated during professional legal services. In Mr. Diljeet Titus, Advocate v. Mr. Alfred A. Adebare & Ors., the Court considered the confidentiality obligations surrounding client-related documentation and the relevance of the BCI Rules and professional privilege.

Therefore, the use of technology must be consistent with the same professional standards that apply to traditional files, emails, physical documents and other methods of handling client information.

---

4. Does Using ChatGPT or Another AI Tool Automatically Violate Confidentiality?

No.

There is an important distinction between:

Using AI as a tool

and

Disclosing confidential information to an AI system without adequate safeguards.

For example, a lawyer may use an AI system to:

• improve grammar in a generic paragraph;
• create a non-confidential checklist;
• brainstorm general legal issues;
• explain a publicly available statutory provision;
• summarise publicly available judgments;
• translate non-confidential text; or
• generate a generic legal research framework.

These uses do not necessarily involve disclosure of privileged client information.

The risk changes significantly when a lawyer uploads:

"Here is my client's confidential agreement. Please identify the breach and advise us on litigation strategy."

The document could contain:

• names;
• addresses;
• financial information;
• commercial terms;
• confidential correspondence;
• litigation strategy;
• personal data; or
• legally privileged material.

The lawyer must then consider where the information is going, how it will be processed, who may access it, how long it may be retained, whether it may be used for model improvement or other purposes, and what contractual and security protections apply.

---

5. The Biggest Risk: Confidential Information Entered Into an External System

One of the most common misconceptions is:

"I am only using AI to help me draft, so confidentiality is not an issue."

The method of processing matters.

When confidential information is entered into an external AI platform, the lawyer should consider whether the information is being transmitted to or processed by a third-party service provider.

This creates several questions:

Where is the data stored?

The lawyer should understand the relevant hosting and storage arrangements.

Who can access it?

Access controls and administrative access matter.

Is the data retained?

The service's retention policy may be relevant.

Is the information used to improve or train models?

The terms applicable to the particular product and account should be examined rather than assumed.

Can the provider use subcontractors?

Cloud and AI services may involve multiple infrastructure providers.

Can the information be deleted?

The lawyer should understand applicable deletion and retention mechanisms.

What happens if there is a security incident?

Incident-response arrangements should be considered.

The answer to these questions can differ substantially between AI products and between consumer and enterprise offerings.

Therefore:

Never assume that every AI tool provides the same level of confidentiality.

---

6. Consumer AI Tools and Enterprise AI Tools Are Not Necessarily Equivalent

Law firms should distinguish between:

consumer-grade AI services

and

enterprise or professionally managed AI environments.

Enterprise products may offer additional contractual, administrative and security controls. Depending on the provider and plan, these may include matters such as:

• organisational accounts;
• access management;
• administrative controls;
• encryption;
• retention controls;
• contractual data-processing terms;
• audit capabilities; and
• restrictions on the use of customer data.

But even an enterprise AI platform should not automatically be treated as legally privileged simply because it is labelled "enterprise."

The lawyer should examine the actual:

• Terms of Service;
• privacy policy;
• data-processing agreement;
• retention policy;
• security documentation;
• data-use provisions; and
• applicable contractual commitments.

---

7. Confidentiality and Data Protection Are Related but Different

This distinction is extremely important.

Legal privilege/confidentiality

This concerns the lawyer's professional obligations toward the client and the legal protection applicable to professional communications.

Data protection

This concerns the processing and protection of personal data under applicable data-protection law.

A document can involve both.

For example, a criminal case file may contain:

• the client's name;
• Aadhaar-related information;
• financial records;
• medical information;
• correspondence;
• witness details; and
• legal advice.

Sending such material to an AI system may therefore raise both professional confidentiality and data-protection questions.

---

8. The Digital Personal Data Protection Framework

India's Digital Personal Data Protection Act, 2023 (DPDP Act) creates a statutory framework for processing digital personal data.

The Act received Presidential assent on 11 August 2023. Its commencement has been notified in phases, rather than all provisions becoming operative simultaneously. The Government notified the DPDP Rules, 2025 on 13 November 2025, with implementation also structured in phases.

The substantive compliance provisions are subject to the notified commencement timeline, with the principal compliance provisions scheduled for the later phase of implementation. Current legal commentary based on the Government's notifications places that major compliance phase at approximately May 2027.

This means law firms should distinguish between:

what is legally operative today

and

what organisations should prepare for under the incoming framework.

A prudent law firm should not wait until the final compliance date before developing secure AI and data-handling practices.

---

9. The Existing Information Technology Framework Also Remains Relevant

The Information Technology Act, 2000 and its related framework have historically imposed obligations concerning protection of sensitive personal data.

Section 43A of the IT Act addresses compensation for failure by a body corporate to implement and maintain reasonable security practices and procedures when handling sensitive personal data or information and wrongful loss or wrongful gain results.

Accordingly, firms handling sensitive information through digital systems should consider appropriate security practices rather than relying solely on confidentiality clauses.

---

10. What About Privilege When AI Processes a Legal Document?

This is one of the most legally sensitive questions.

There is currently no Indian Supreme Court ruling establishing a blanket rule that uploading privileged material to an AI platform automatically destroys privilege.

Nor is there a general statutory provision specifically stating that use of an AI system constitutes waiver of advocate-client privilege.

Therefore, it would be incorrect to state categorically that:

"Using AI automatically waives privilege."

That proposition is not presently established as a general rule of Indian law.

However, the absence of such a ruling does not make careless disclosure safe.

A lawyer should consider whether sending confidential information to an external third party is consistent with:

• the client's instructions;
• the advocate's professional obligations;
• applicable privilege rules;
• contractual confidentiality obligations;
• data-protection requirements;
• court directions; and
• the terms governing the AI service.

The prudent approach is to minimise unnecessary disclosure.

---

11. Client Consent: Is It Enough?

Client consent can be relevant, but it should not be treated as a complete solution to every AI-related risk.

Section 132 of the BSA expressly refers to the client's express consent in relation to disclosure of protected professional communications.

However, obtaining a generic statement such as:

"The client agrees that the lawyer may use AI"

does not necessarily resolve every issue.

The firm should still consider:

• what information is being processed;
• for what purpose;
• which AI provider is being used;
• whether the provider retains the information;
• whether the information may be used for model development;
• security controls;
• cross-border processing;
• applicable law;
• third-party access;
• confidentiality undertakings; and
• the client's specific instructions.

Consent should therefore be informed and appropriately scoped, where consent is relied upon.

---

12. Can Lawyers Use AI for Legal Research?

Yes, but legal research creates a different risk.

A lawyer can use AI to:

• identify potential cases;
• generate research questions;
• summarise legal principles;
• locate relevant provisions;
• compare arguments; and
• structure a research memorandum.

But AI-generated legal research must be independently verified.

This is particularly important because generative AI systems can produce:

• incorrect citations;
• non-existent judgments;
• inaccurate quotations;
• outdated law;
• incomplete statutory analysis; or
• incorrect descriptions of judicial decisions.

The responsibility for legal advice remains with the lawyer.

A practical rule is:

AI may assist legal research; it should not replace legal verification.

Every authority relied upon in a pleading, opinion or submission should be checked against an authoritative source.

---

13. Can Lawyers Upload a Complete Case File to AI?

As a general risk-management principle, lawyers should avoid uploading complete confidential case files to a general-purpose external AI service unless the firm's legal, contractual, security and client-consent framework has been appropriately assessed.

A safer workflow may involve:

Instead of uploading the entire file:

Step 1: Remove unnecessary personal information.

Step 2: Anonymise names and identifiers where possible.

Step 3: Remove privileged material that the AI does not need.

Step 4: Provide only the minimum information required for the task.

Step 5: Use an appropriately secured and contractually reviewed AI environment.

Step 6: Independently verify the AI output.

This follows a basic data-governance principle:

If the AI does not need the information, do not give it the information.

---

14. Data Minimisation Should Become a Law-Firm AI Principle

A law firm should ask:

What is the minimum information necessary for the AI to perform this task?

For example, instead of:

"Review this entire 100-page client agreement and tell me whether our client has breached the agreement."

A lawyer may first anonymise the document and provide only the relevant provisions if the task permits.

Instead of:

"Here is my client's entire criminal case file."

The lawyer could ask an AI system a general question:

"What legal issues should an advocate examine when analysing a prosecution based on circumstantial evidence?"

The second approach does not expose client information unnecessarily.

---

15. AI Should Not Become an Unauthorised Member of the Legal Team

One useful way of thinking about AI is:

Treat an external AI service as a third-party service provider—not as a member of your chambers.

Before allowing confidential information to be processed, the firm should ask:

• Has the service been approved?
• Who controls the account?
• Where is information processed?
• What happens to uploaded data?
• What contractual protections exist?
• Who can access the account?
• Is multi-factor authentication enabled?
• What is the retention period?
• Can data be deleted?
• Is there an audit trail?
• What happens after the engagement ends?

This is particularly important for firms handling high-value corporate, criminal, regulatory, arbitration and constitutional matters.

---

16. A Law Firm Should Have an AI Policy

Every modern law firm should consider implementing an internal AI Use and Confidentiality Policy.

It can establish:

Permitted Uses

Examples:

• public legal research;
• grammar and formatting;
• generic drafting assistance;
• non-confidential brainstorming;
• summarisation of publicly available material.

Restricted Uses

Examples:

• confidential client documents;
• privileged communications;
• personal data;
• investigation material;
• trade secrets;
• unpublished settlement terms;
• sensitive corporate information.

Prohibited Uses

Depending on the firm's risk assessment:

• uploading confidential material to unapproved AI platforms;
• entering privileged communications into consumer accounts;
• sharing passwords or credentials;
• using AI-generated authorities without verification;
• allowing AI to make unsupervised legal decisions.

---

17. Lawyers Should Maintain Human Oversight

AI should not become the final decision-maker in legal practice.

For example, AI can help identify possible arguments.

But the advocate must determine:

• whether the argument is legally sound;
• whether the authority remains good law;
• whether facts support the argument;
• whether the argument is ethically permissible;
• whether disclosure is appropriate;
• and whether it should actually be presented to the Court.

The lawyer remains accountable for the professional work product.

---

18. Court Filings Require Particular Caution

The risks become even greater when AI-generated material is incorporated into:

• pleadings;
• affidavits;
• written submissions;
• legal opinions;
• applications;
• arbitration submissions; or
• other documents filed before a court or tribunal.

An advocate should verify:

Every citation.
Every quotation.
Every statutory provision.
Every factual assertion.
Every case reference.

AI-generated material should never be treated as authoritative merely because it sounds legally sophisticated.

---

19. What If the AI Generates a False Case Citation?

This is one of the most obvious practical risks.

An AI system may produce a citation that appears authentic but does not exist or may inaccurately describe a real decision.

A lawyer who relies upon it without verification risks:

• submitting incorrect material to the Court;
• misleading the client;
• undermining the credibility of the pleading;
• wasting judicial time; and
• potentially facing professional consequences depending on the circumstances.

The solution is straightforward:

AI-generated legal research must be independently checked against authoritative legal databases, official court websites, legislation and other reliable sources before reliance.

---

20. What Lawyers Should Never Put Into an Unapproved AI Tool

As a practical risk-control measure, law firms should generally prohibit or tightly restrict uploading the following to unapproved systems:

• client passwords;
• bank account credentials;
• Aadhaar/PAN details;
• medical records;
• confidential corporate agreements;
• unpublished M&A documents;
• privileged correspondence;
• litigation strategy;
• investigation files;
• witness statements;
• settlement negotiations;
• trade secrets;
• source code belonging to clients;
• confidential government documents; and
• documents subject to specific court confidentiality directions.

The exact policy should depend upon the firm's practice areas and contractual obligations.

---

21. What About Client Documents Already in Cloud Storage?

Cloud storage itself is not necessarily incompatible with confidentiality.

Modern law firms routinely use:

• cloud document-management systems;
• email platforms;
• case-management systems;
• electronic discovery platforms; and
• remote collaboration tools.

The legal question is not simply:

"Is the document in the cloud?"

The relevant questions include:

Who controls the system?
What contractual safeguards exist?
What security measures are implemented?
Who can access the information?
Where is it processed?
What happens in the event of a breach?

The same approach should be applied to AI platforms.

---

22. A Practical AI Safety Framework for Law Firms

EOS Chambers—or any law firm considering AI adoption—can use a simple 7-point AI confidentiality framework:

1. Classify

Identify whether the information is public, internal, confidential, sensitive or privileged.

2. Minimise

Provide only the information necessary for the task.

3. Anonymise

Remove names, identifiers and unnecessary personal information wherever possible.

4. Verify

Review the AI provider's terms, privacy policy, security documentation and data-use practices.

5. Authorise

Use only AI tools approved by the firm.

6. Verify the Output

Independently check legal authorities, facts, quotations and analysis.

7. Maintain Human Control

The final legal judgment must remain with the advocate.

---

23. A Simple Risk Matrix

This is a risk-management framework, not a statement that Indian law assigns these exact categories or risk levels.

---

24. Can Lawyers Safely Use AI?

The answer is:

Yes—but not casually.

AI can be a useful professional tool when deployed within an appropriate framework.

The issue is not whether lawyers should use technology.

The issue is whether technology is being used in a way that remains consistent with:

• professional confidentiality;
• legal privilege;
• client instructions;
• data-protection obligations;
• information-security standards;
• court requirements; and
• professional judgment.

Indian law already places confidentiality obligations upon advocates. The arrival of AI does not remove them.

---

25. The Future of Legal Practice Will Be Human + AI

Artificial Intelligence will probably become increasingly integrated into legal practice.

Lawyers may use AI for:

• first-level document review;
• legal research;
• contract analysis;
• chronology preparation;
• translation;
• discovery;
• drafting assistance;
• due diligence; and
• knowledge management.

But the core responsibilities of an advocate remain human:

judgment, ethics, confidentiality, accountability and responsibility to the client and the Court.

AI can process information.

It cannot take over the advocate's professional responsibility.

---

Conclusion

The question is no longer whether lawyers will use Artificial Intelligence.

They already are.

The more important question is how responsibly they use it.

Indian law provides a strong framework protecting professional communications between advocates and clients. Section 132 of the Bharatiya Sakshya Adhiniyam, 2023 protects specified professional communications, documents and legal advice, subject to statutory exceptions. The Bar Council of India Rules separately require advocates not to breach their professional obligations concerning such communications.

At the same time, India's digital privacy framework is evolving, with the DPDP Act, 2023 and DPDP Rules, 2025 being implemented through a phased commencement framework.

Therefore, the responsible approach is neither:

"Never use AI."

nor:

"AI is safe for everything."

It is:

Use AI where it creates value, but never outsource professional responsibility.

For a law firm, the most important principle should remain simple:

Technology may assist the lawyer. It must never compromise the client's trust.