AI in Legal Work: What Is Safe for Research and Drafting

AI for lawyers is useful for legal research summaries, contract clause drafting, and document review preparation. It is unreliable for case citation accuracy, unsuitable for tasks requiring client confidentiality protections, and dangerous when used without verification. Here is an honest task-by-task assessment for Kenyan legal practitioners.

The Two Non-Negotiable Risks

Before discussing what AI can do in legal practice, we need to address the two risks that override everything else.

Hallucination: AI Invents Case Law

AI language models generate text that sounds authoritative regardless of whether it is accurate. In legal work, this manifests as fabricated case citations. An AI tool asked to "cite Kenyan case law supporting the right to a fair hearing" may produce case names, year references, and even plausible-sounding quotes from judgments that do not exist.

This has already caused professional embarrassment internationally. Lawyers in the US have been sanctioned for submitting AI-generated briefs containing entirely fictitious case citations. In Kenya, submitting fabricated authority to the High Court or any tribunal would violate the Advocates Act's professional conduct standards and could result in disciplinary proceedings before the Law Society of Kenya.

The rule is simple: every case, statute, and legal principle an AI produces must be verified against primary sources (Kenya Law Reports, the National Council for Law Reporting database, or the official Kenya Gazette). If you cannot find the cited authority in a recognized source, it does not exist.

Confidentiality: Client Data in AI Tools

Typing client details, case facts, or privileged communications into a public AI tool creates a confidentiality risk. Free tiers of most AI platforms use conversations for model training or store them on servers outside your control. This potentially breaches advocate-client privilege under the Advocates Act and the Evidence Act.

Practical safeguards: never enter client names, case numbers, or identifying details into free public AI tools. If you use AI for drafting, anonymize the facts first. Replace names with placeholders ("Party A," "the Defendant"), remove specific amounts, and strip identifying details. Enterprise-tier AI tools with contractual data privacy commitments are safer for firms that can afford them.

Tasks Where AI Genuinely Helps

With those guardrails established, here is where AI adds value in Kenyan legal practice.

Legal Research: Starting Points, Not Endpoints

AI is useful for generating a research starting framework. Ask it to outline the legal principles governing, say, landlord-tenant disputes in Kenya. The AI will identify relevant statutes (the Landlord and Tenant Act, the Rent Restriction Act), point to relevant constitutional provisions, and outline the general legal framework.

This starting framework saves time. Instead of beginning research from scratch, you start with a structured outline that you then verify and deepen using primary sources. The AI's output is a hypothesis to test, not an authority to cite.

For practitioners who handle diverse matters (common in smaller Kenyan firms where an advocate might deal with conveyancing, family law, and commercial disputes in the same week), AI provides a quick refresher on areas you practice less frequently.

Contract and Document Drafting

AI excels at producing first drafts of standard legal documents: non-disclosure agreements, employment contracts, lease agreements, demand letters, and commercial agreements. Provide the key terms (parties, obligations, term, jurisdiction, governing law) and the AI generates a structured draft.

For Kenyan-specific documents, you need to verify that the AI includes the correct legal formalities. A Kenyan lease agreement must comply with the Registered Land Act provisions (for registered land) and any applicable county bylaws. An employment contract must reflect the Employment Act's minimum requirements for leave days, termination notice, and other statutory provisions.

The AI draft gets you a majority to a majority of the way. Your legal expertise fills the remaining a notable share to a significant portion: ensuring compliance with Kenyan law, adding clauses specific to the client's situation, and removing provisions that do not apply in this jurisdiction.

Summarizing Lengthy Documents

Reviewing a 200-page due diligence report, a complex trust deed, or a voluminous court record is time-intensive. AI can summarize these documents, extracting key clauses, obligations, dates, and potential issues.

Upload the document (or paste sections, with identifying details removed) and ask: "Summarize the key obligations of each party. Flag any clauses that contain unusual terms, indemnity provisions, or penalty clauses."

The summary gives you a map of the document. You then read the flagged sections in detail rather than reading every page with equal attention.

Client Communication Drafting

Writing letters to clients explaining legal positions, updating them on case progress, or outlining their options is routine but time-consuming. AI drafts these communications quickly when you provide the substance.

"Draft a letter to a client explaining that their employment dispute has been referred to the Employment and Labour Relations Court. The first mention date is [date]. Explain what they should expect at the mention and what documents they need to prepare."

Review the draft for accuracy, adjust the tone to match your client relationship, and ensure it reflects the specific procedural requirements of the ELRC.

Tasks Where AI Is Unsafe or Unreliable

Case Citation and Authority

As discussed above, AI fabricates legal citations. Do not rely on AI-generated case references without verifying each one in the Kenya Law Reports or equivalent primary sources. This applies to statutes as well: AI sometimes cites repealed legislation or misidentifies section numbers.

Legal Advice and Opinion

AI does not understand the specific circumstances of a client's situation the way an advocate does. It cannot assess credibility of witnesses, gauge the likely approach of a particular judge, or weigh the practical considerations (cost, time, relationship preservation) that inform legal strategy. AI can outline the law. It cannot advise on what to do.

Court Submissions

Any document filed in court must be the work product of the advocate on record. Using AI to draft is acceptable (advocates have always used templates, precedent files, and research assistants). But the advocate must review every word, verify every citation, and take professional responsibility for the content. Filing an AI draft without thorough review is professionally reckless.

Matters Involving Sensitive or Privileged Information

Disputes involving domestic violence, child custody, sexual offences, or other sensitive matters require extreme care. Entering case details into AI tools, even anonymized, risks re-identification if the facts are distinctive enough. For these matters, keep AI out of the substantive work entirely.

Practical Workflow for a Kenyan Law Firm

A firm wanting to introduce AI into its practice can follow this graduated approach:

Month 1: Research assistance only. Use AI to generate research outlines and frameworks. Verify everything against primary sources. Track time savings.

Month 2: Add standard document drafting. Use AI to draft standard agreements and letters for routine matters. Senior review before any document goes to a client or court.

Month 3: Expand to document review. Use AI to summarize lengthy documents. Cross-check AI summaries against manual review for the first few documents to calibrate accuracy.

Ongoing: Never use AI for citations without verification, privileged matters without anonymization, or court submissions without thorough advocate review.

This graduated approach builds confidence and catches errors early, before they reach clients or courts.

The Professional Responsibility Dimension

The Law Society of Kenya has not issued specific guidance on AI use in legal practice as of 2026. However, existing professional conduct rules apply. Advocates owe duties of competence, diligence, and confidentiality. Using AI in a way that compromises any of these duties exposes the advocate to disciplinary action.

Competence means understanding the limitations of the tools you use. If you use AI and do not know about its tendency to hallucinate citations, you are not using it competently.

Diligence means verifying AI output. Treating an AI draft as final without review fails the diligence standard.

Confidentiality means protecting client information from unauthorized disclosure. Entering client data into insecure AI platforms breaches this duty.

The advocate who uses AI responsibly (with verification, anonymization, and professional oversight) is practising more efficiently. The advocate who uses AI carelessly is creating professional risk.

For legal professionals interested in building practical AI skills with proper safeguards, the free welcome module of our AI and Automation course covers the fundamentals of working with AI tools effectively and safely.

FAQ

Can I use AI to prepare for court appearances?

Yes, for preparation tasks: summarizing case files, outlining arguments, identifying potential counter-arguments, and drafting examination questions. The preparation is yours. The AI helps organize your thinking. Do not rely on AI for real-time assistance during proceedings, and verify every legal authority it suggests.

Which AI tool is safest for legal work?

Enterprise tiers of Claude (Anthropic) and ChatGPT (OpenAI) offer contractual data privacy commitments and do not use your inputs for training. These are significantly safer than free tiers for legal work. Some legal-specific AI tools (like CoCounsel) are designed with legal confidentiality requirements in mind. Evaluate any tool's data handling policy before using it with client-adjacent information.

Will AI replace paralegals and junior advocates?

AI will change the work, not eliminate the roles. Research, drafting, and document review will be faster with AI assistance. But someone still needs to verify the AI's output, apply professional judgment, interact with clients, and take responsibility for the work product. Junior advocates and paralegals who develop AI skills will handle larger workloads more efficiently. Those skills make them more valuable, not redundant.

How do I explain AI use to clients?

Transparency is the best approach. Explain that your firm uses AI tools to improve efficiency in research and drafting, that all AI-generated work is reviewed by a qualified advocate, and that client confidentiality protections are maintained through anonymization and secure tools. Most clients appreciate efficiency, provided quality and confidentiality are not compromised.