Use of AI Tools in Document Review

Introduction

AI tools in document review refer to the use of artificial intelligence, machine learning, natural language processing (NLP), generative AI, and related technologies to examine, classify, summarise, compare, and extract information from large volumes of documents.

In legal practice, AI-assisted document review is increasingly relevant in:

• litigation and e-discovery;
• contract review;
• due diligence;
• compliance investigations;
• employment records;
• regulatory audits;
• legal research;
• document classification;
• identification of privileged or confidential material.

AI can reduce the amount of manual review required, but human supervision remains important, particularly where documents contain confidential, privileged, personal, or legally sensitive information.

1. Meaning of AI-Assisted Document Review

Traditional document review involves lawyers or employees manually examining documents to determine:

• relevance;
• confidentiality;
• privilege;
• legal significance;
• factual importance;
• compliance issues.

AI-assisted review uses algorithms to assist with these tasks.

For example, an AI system may examine 100,000 emails and identify documents containing terms relating to a particular dispute, classify documents according to relevance, or summarise lengthy correspondence.

Common technologies include:

Natural Language Processing

NLP enables software to analyse human language and identify:

• names;
• dates;
• organisations;
• legal concepts;
• relationships between documents;
• relevant phrases.

Machine Learning

Machine-learning systems can learn from documents classified by human reviewers and apply similar classifications to other documents.

Generative AI

Generative AI can:

• summarise documents;
• compare contracts;
• identify clauses;
• draft document summaries;
• answer questions based on document collections.

2. Uses of AI in Document Review

A. Document Classification

AI can classify documents into categories such as:

• relevant;
• irrelevant;
• privileged;
• confidential;
• potentially responsive.

This is particularly useful during litigation involving thousands or millions of documents.

B. Contract Review

AI can identify important contractual provisions such as:

• termination clauses;
• indemnity clauses;
• confidentiality provisions;
• arbitration clauses;
• limitation-of-liability clauses;
• non-compete provisions;
• governing-law clauses.

It can also compare a contract against a standard template and highlight deviations.

C. Due Diligence

During mergers and acquisitions, lawyers may have to review thousands of documents.

AI can assist in identifying:

• material contracts;
• litigation;
• regulatory issues;
• intellectual-property documents;
• employment agreements;
• financial obligations.

D. E-Discovery

AI is particularly relevant to electronic discovery.

It can assist in reviewing:

• emails;
• instant messages;
• PDFs;
• spreadsheets;
• electronic documents;
• databases.

Technology-assisted review can help lawyers prioritise documents that are more likely to be relevant.

3. Advantages of AI-Assisted Document Review

Speed

AI can process large quantities of documents much faster than manual review.

Cost Reduction

Automated review can reduce the number of hours required for repetitive document-analysis tasks.

Consistency

AI can apply the same classification criteria across large document collections.

Search Capability

AI can identify patterns and concepts that may not be discovered through simple keyword searches.

Summarisation

Generative AI can provide concise summaries of lengthy documents.

4. Legal Risks of Using AI for Document Review

A. Confidentiality

Legal documents frequently contain confidential information.

Uploading such documents to an inappropriate external AI service may create confidentiality risks.

Lawyers must therefore consider:

• where data is stored;
• who can access it;
• whether data is retained;
• whether it is used for model training;
• contractual safeguards;
• cybersecurity protections.

B. Legal Professional Privilege

Documents containing privileged communications require special protection.

AI systems should not be allowed to inadvertently expose privileged material to unauthorised persons.

C. Hallucinations

Generative AI can produce incorrect information that appears convincing.

For example, an AI system could:

• invent a case;
• misstate a contractual clause;
• misunderstand a legal provision;
• attribute a statement to the wrong document.

Therefore, AI-generated results should be verified against the original documents.

D. Bias

AI systems can reproduce biases present in their training data or classification processes.

This can become problematic where AI is used to classify documents concerning:

• employees;
• discrimination claims;
• disciplinary proceedings;
• recruitment;
• workplace complaints.

E. Data Protection

Documents may contain personal information such as:

• names;
• addresses;
• salary information;
• medical information;
• identification numbers.

AI-assisted review must therefore be designed consistently with applicable data-protection requirements.

5. Human Oversight

AI should generally function as an assistance tool rather than an unquestioned substitute for legal judgment.

A lawyer should verify:

1. important AI-generated conclusions;
2. citations;
3. extracted contractual provisions;
4. privilege classifications;
5. summaries of important documents;
6. documents identified as potentially relevant.

A useful workflow is:

AI processing → Human review → Verification → Final legal decision

6. Important Case Laws

1. Da Silva Moore v. Publicis Groupe (2012)

This United States case is one of the important early decisions concerning technology-assisted review (TAR) in electronic discovery.

The court accepted the use of predictive coding/TAR as a method for assisting document review.

Importance

The case demonstrated that courts could recognise sophisticated technology-assisted methods as legitimate tools for large-scale discovery.

Principle: Technology can assist lawyers in reviewing large document collections, provided the process is appropriately managed and transparent.

2. Rio Tinto PLC v. Vale S.A. (2015)

The United States District Court discussed the use of predictive coding for electronic discovery.

The case is significant because it recognised the practical usefulness of technology-assisted review in handling large quantities of electronically stored information.

Importance

The decision illustrates the growing judicial acceptance of machine-learning-assisted document review.

3. Pyrrho Investments Ltd v MWB Property Ltd (2016)

The English High Court considered the use of predictive coding in electronic disclosure.

The court permitted the use of predictive coding for document review.

Importance

The case is significant in demonstrating that AI-assisted review can be compatible with judicial disclosure obligations when appropriately used.

4. Brown v. BCA Trading Ltd (2021)

This English case concerned issues surrounding electronic disclosure and the management of large volumes of documents.

It illustrates the increasing importance of proportionate and technologically informed approaches to disclosure.

Importance

Modern litigation requires parties to adopt appropriate methods for managing large electronic document collections rather than relying exclusively on manual review.

5. State of Wisconsin v. Loomis (2016)

This case concerned the use of an algorithmic risk-assessment system in the criminal justice context rather than ordinary document review.

The Wisconsin Supreme Court considered concerns surrounding the use of proprietary algorithms in judicial decision-making.

Importance for AI document review

The case demonstrates a broader legal principle: when important legal consequences depend upon algorithmic systems, transparency, accuracy, and appropriate human oversight become significant concerns.

6. Mata v. Avianca, Inc. (2023)

This is a particularly important modern case involving generative AI and legal practice.

Lawyers used ChatGPT in preparing a court filing, and the filing contained fictitious case authorities generated by the AI system.

The court imposed sanctions.

Importance

The case illustrates that lawyers remain responsible for verifying AI-generated legal material.

Principle: A lawyer cannot avoid professional responsibility merely because an inaccurate statement was generated by an AI system.

7. In re Valsartan, Losartan, and Irbesartan Products Liability Litigation (2022)

The litigation involved issues concerning electronic discovery and the use of advanced technological methods for document review.

Importance

The broader significance of such e-discovery cases is that technology must be integrated into a defensible discovery methodology, with appropriate validation and quality control.

7. AI and Privileged Documents

One of the most sensitive issues is privilege review.

Suppose a company has 500,000 emails and 20,000 contain communications with lawyers.

An AI system may help identify potentially privileged communications, but the final privilege determination should generally be subject to appropriate legal review.

A mistake could result in:

• waiver of privilege;
• disclosure of confidential information;
• procedural disputes;
• sanctions.

Therefore, organisations should establish clear privilege-review protocols.

8. AI in Employment Document Review

AI can also be used to review employment-related documents.

For example, an organisation could use AI to identify:

• employment contracts;
• disciplinary records;
• workplace complaints;
• wage records;
• termination documents;
• leave records;
• workplace policies.

However, employee information can contain highly sensitive personal data.

Accordingly, employers should consider:

• data minimisation;
• access controls;
• confidentiality;
• accuracy;
• retention periods;
• employee privacy;
• human review.

9. Best Practices

Organisations using AI for document review should establish an AI governance framework.

Important safeguards include:

1. Verify AI Outputs

Never rely blindly on AI-generated summaries or legal conclusions.

2. Protect Confidential Information

Use appropriate security controls and approved systems.

3. Maintain Audit Trails

Keep records of:

• what AI system was used;
• what documents were processed;
• what instructions were given;
• what outputs were generated;
• who reviewed the results.

4. Human-in-the-Loop Review

Important legal decisions should receive human oversight.

5. Test Accuracy

The AI system should be tested against a sample of documents reviewed by humans.

6. Control Access

Only authorised personnel should have access to sensitive document collections.

7. Review AI Contracts

Organisations should understand:

• data ownership;
• retention;
• confidentiality;
• model-training provisions;
• security responsibilities.

10. AI Document Review and Indian Legal Practice

India does not currently have a single comprehensive statute specifically governing AI-assisted legal document review.

The legal framework therefore comes from a combination of:

• constitutional rights;
• data-protection law;
• contract law;
• professional obligations;
• evidence law;
• procedural rules;
• confidentiality and privilege principles.

The Digital Personal Data Protection Act, 2023 is particularly relevant where AI document-review systems process personal data.

The constitutional right to privacy recognised in K.S. Puttaswamy v. Union of India (2017) is also important when AI systems process personal information.

Conclusion

AI tools can significantly transform document review by making it faster, scalable and more efficient. They are particularly useful for e-discovery, contract analysis, due diligence, compliance reviews and large-scale document classification.

However, AI-generated results can contain errors, hallucinations and classification mistakes. Confidentiality, privilege, privacy and cybersecurity also create significant legal concerns.

The case law, particularly Da Silva Moore, Rio Tinto, Pyrrho Investments and Mata v. Avianca, demonstrates the developing judicial approach to technology-assisted legal work.

The central principle is that AI can assist the legal professional, but responsibility for the final legal judgment remains with the human reviewer.