Use of AI in Document Review

1. Introduction

AI in document review refers to the use of artificial intelligence, machine learning, natural-language processing (NLP), and related technologies to examine large volumes of documents and identify information relevant to a legal matter, investigation, compliance exercise, audit, or employment dispute.

In legal practice, AI-assisted document review may be used for:

• identifying relevant documents;
• classifying documents;
• detecting duplicates;
• identifying privileged material;
• extracting names, dates and clauses;
• searching emails and communications;
• identifying potentially responsive documents;
• summarising lengthy documents;
• detecting patterns across large datasets; and
• assisting lawyers during electronic discovery (e-discovery).

AI does not automatically replace the lawyer's professional judgment. Its output normally requires human verification, particularly where privilege, confidentiality, relevance or legal conclusions are involved.

2. How AI Document Review Works

A typical AI-assisted review process can involve the following stages:

A. Collection

Electronic information is collected from sources such as:

• emails;
• cloud storage;
• company databases;
• messaging systems;
• contracts;
• HR records;
• accounting records; and
• litigation files.

B. Processing

The documents may be converted into searchable formats and analysed for:

• metadata;
• dates;
• authors;
• recipients;
• keywords;
• document types; and
• duplicate copies.

C. AI Classification

Machine-learning systems can classify documents according to categories such as:

• relevant;
• irrelevant;
• potentially privileged;
• confidential;
• responsive to a particular request; or
• requiring human review.

D. Human Review

Lawyers or trained reviewers examine documents identified by the system.

E. Quality Control

The review team tests the AI system's results to identify:

• false positives;
• false negatives;
• inconsistent classifications; and
• missed relevant documents.

3. Technology Used in AI Document Review

Natural Language Processing

NLP enables systems to analyse human language and identify relationships between words and concepts.

Machine Learning

Machine-learning models can learn from documents previously classified by reviewers and use that information to classify additional documents.

Predictive Coding

Predictive coding uses machine learning to identify documents that are likely to be relevant to a particular legal issue.

Generative AI

Modern generative-AI systems can assist with:

• summarisation;
• chronology creation;
• clause comparison;
• issue identification;
• extracting information from documents; and
• drafting preliminary review notes.

The final legal determination should remain subject to appropriate human review.

4. Advantages of AI Document Review

A. Speed

AI can process very large document collections substantially faster than manual review.

B. Cost Reduction

Automation may reduce the amount of routine manual review required.

C. Consistency

A properly designed system can apply the same classification criteria across large datasets.

D. Detection of Relevant Information

AI can identify relationships and patterns that may be difficult to locate through simple keyword searches.

E. E-Discovery

AI is particularly useful where litigation involves thousands or millions of electronic documents.

5. Legal Risks Associated with AI Document Review

A. Confidentiality

Legal documents frequently contain sensitive information.

Uploading confidential documents into an inappropriate AI system may create confidentiality and data-protection risks.

Lawyers must therefore consider:

• where the data is stored;
• who can access it;
• whether the provider retains the data;
• whether it is used for model training;
• encryption;
• access controls; and
• contractual protections.

B. Privilege

A document may contain attorney-client communications or attorney work product.

AI tools may incorrectly classify privileged documents as non-privileged.

Therefore, privilege review should not be delegated blindly to an AI system.

C. Hallucinations

Generative AI can produce inaccurate information or invent references that do not exist.

For document review, this is particularly dangerous because an incorrect summary can affect litigation strategy.

D. Bias

AI systems can reproduce biases contained in training data or introduced through the design of the review process.

A review system should therefore be tested and monitored.

E. Data Protection

Personal data contained in documents may trigger privacy and data-protection obligations.

This is particularly important for:

• employee records;
• medical information;
• financial information;
• identification documents;
• customer data; and
• communications.

6. AI and Legal Professional Responsibility

A lawyer using AI remains responsible for the legal work product.

AI should generally be treated as an assistive tool rather than an independent legal decision-maker.

A lawyer should consider:

1. whether the tool is appropriate for the matter;
2. whether confidential information can safely be entered;
3. whether outputs are accurate;
4. whether important documents have been missed;
5. whether privilege has been protected;
6. whether the result has been independently verified.

7. Important Case Laws

There is no single Indian Supreme Court judgment that comprehensively establishes a legal framework specifically governing AI-assisted document review. The following cases are nevertheless highly relevant because they establish principles concerning electronic evidence, privacy, confidentiality, professional responsibility and technology.

1. Anvar P.V. v. P.K. Basheer (2014)

The Supreme Court dealt with the admissibility of electronic records under the Indian Evidence Act.

The judgment established important principles regarding the proof and admissibility of electronic evidence.

Relevance to AI document review

AI review commonly involves emails, electronic files and other digital records. Therefore, merely identifying a document through AI does not automatically establish its evidentiary admissibility.

The underlying electronic record must still satisfy applicable evidentiary requirements.

2. Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal (2020)

The Supreme Court revisited the requirements concerning electronic evidence and Section 65B certification of the Indian Evidence Act.

The judgment clarified important aspects of the admissibility of electronic records.

Relevance

AI systems may identify or organise electronic evidence, but the AI-generated classification or summary cannot replace statutory requirements governing the authenticity and admissibility of the underlying electronic record.

3. K.S. Puttaswamy v. Union of India (2017)

The Supreme Court recognised privacy as a constitutionally protected fundamental right under Article 21 and the broader constitutional framework.

The Court discussed principles including:

• privacy;
• dignity;
• autonomy;
• informational privacy; and
• proportionality.

Relevance

AI document-review systems may process enormous amounts of personal information.

Where documents contain employee or customer data, privacy considerations become particularly important. AI-assisted review should therefore incorporate appropriate safeguards concerning collection, processing and access to personal information.

4. R. Rajagopal v. State of Tamil Nadu (1994)

The Supreme Court considered important principles concerning privacy and publication of personal information.

The case is relevant to the broader legal principle that individuals have interests in controlling the disclosure of private information.

Relevance

AI document review can expose sensitive personal information contained in documents. Organisations therefore need appropriate confidentiality and privacy controls when using AI to process such material.

5. State of Punjab v. Baldev Singh (1999)

The Supreme Court emphasised the importance of procedural safeguards in relation to searches and evidence.

Relevance

AI-based document review may be conducted in the context of investigations or discovery exercises. The use of technology does not eliminate the need to comply with applicable procedural and evidentiary safeguards.

6. Sahara India Real Estate Corporation Ltd. v. SEBI (2012)

The Supreme Court considered issues concerning confidentiality and access to sensitive information in judicial proceedings.

The judgment recognised the need to balance transparency in judicial proceedings with legitimate confidentiality concerns.

Relevance

Large-scale AI document review can involve highly confidential corporate and personal information. The case is relevant to the broader principle that sensitive information may require appropriate protection even within legal proceedings.

7. Shreya Singhal v. Union of India (2015)

The Supreme Court considered constitutional issues concerning online speech and intermediary-related regulation.

Although the case was not about AI document review, it is relevant to the broader legal environment governing digital information and online platforms.

Relevance

AI review increasingly operates on digitally stored communications. Organisations must therefore understand that technological processing does not remove constitutional and statutory protections applicable to digital information.

8. AI Document Review in Employment Law

AI-assisted review is increasingly relevant to employment disputes.

For example, an employer may use AI to review:

• employee emails;
• disciplinary records;
• employment contracts;
• attendance records;
• performance communications;
• internal complaints;
• workplace investigation documents.

However, employers should consider whether the review complies with applicable:

• privacy requirements;
• employment contracts;
• company policies;
• confidentiality obligations;
• data-protection laws; and
• procedural fairness requirements.

9. AI and Privileged Documents

One of the most important issues is legal professional privilege.

Suppose a company gives an AI system access to 500,000 emails. Among them are:

• communications between the company and its lawyers;
• litigation strategy;
• legal advice; and
• internal discussions about the advice.

The AI system could accidentally classify privileged material as ordinary business communication.

A strong review process should therefore include:

AI screening → privilege filters → human privilege review → quality control.

10. Human Oversight

Human supervision is especially important when AI is used for:

• privilege determinations;
• final relevance decisions;
• legal conclusions;
• court submissions;
• regulatory responses;
• employee disciplinary decisions.

AI can help identify information, but the lawyer or responsible professional should retain responsibility for the final legal judgment.

11. Best Practices for AI Document Review

Organisations should consider adopting an AI document-review policy containing:

1. Data classification

Identify whether documents contain confidential, privileged or personal information.

2. Approved AI tools

Only approved systems should be used for sensitive legal documents.

3. Access controls

Limit access to authorised employees and lawyers.

4. Human verification

Important AI-generated conclusions should be reviewed by qualified personnel.

5. Audit trails

Maintain records showing how documents were processed and classified.

6. Accuracy testing

Test the system for false positives and false negatives.

7. Privilege protection

Use dedicated procedures for identifying and protecting privileged communications.

8. Data retention

Documents should not be retained by an AI provider longer than necessary or permitted.

12. AI Document Review and Indian Law

The legal framework must be considered alongside India's developing digital and data-protection regime.

Relevant legal areas include:

• constitutional privacy;
• electronic evidence;
• information technology law;
• data protection;
• confidentiality;
• legal professional privilege;
• employment law; and
• procedural law.

The Digital Personal Data Protection Act, 2023 is particularly relevant where AI systems process personal data, although the precise obligations depend on the applicable provisions, rules and circumstances.

13. Conclusion

AI-assisted document review can significantly improve the speed and scalability of legal document analysis. It can help lawyers identify relevant documents, organise electronic evidence, detect patterns and reduce repetitive manual work.

However, AI does not eliminate legal responsibility. Confidentiality, privilege, privacy, accuracy, electronic-evidence requirements and human oversight remain essential.

Cases such as Anvar P.V. v. P.K. Basheer, Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, K.S. Puttaswamy v. Union of India, and R. Rajagopal v. State of Tamil Nadu provide important legal principles for understanding the treatment of electronic information, privacy and evidence in an AI-assisted legal environment.

The emerging approach is therefore best understood as AI-assisted document review with accountable human supervision, rather than completely autonomous legal review.