AI for Law Firms in Kenya: How Do You Use It Safely?

TL;DR:

• Lawyers in Kenya can use AI for low-risk work: understanding a public judgment, planning research, tidying non-confidential text and testing an argument. Every output still needs a lawyer's check.
• Court filings are the danger zone. Kenyan judges have called out fictitious citations, and in 2026 the High Court struck out or set aside work it found to be machine-generated, even where no fake case was identified.
• The Judiciary's AI policy and Practice Directions were still drafts when we checked on 8 November 2026. We found no published Law Society of Kenya (LSK) AI guidance for advocates. Your existing duties apply in full.
• Client information stays out of public AI tools. Section 134 of the Evidence Act, the LSK's 2017 Code and the Data Protection Act 2019 all point the same way.
• Open every authority on Kenya Law and read it yourself before relying on it. A real case can still be the wrong case.

Who this guide is for

This guide is for advocates and partners in Kenyan firms, pupils and paralegals, in-house counsel, compliance officers and law students.

You do not need any technical background. If you read judgments, draft documents or advise people on the law, this is for you.

This is general information from a training company. It is not legal advice. FuKazee teaches practical AI skills, and we are not advocates. For a decision on a live matter, speak to a qualified advocate and read the current rules yourself.

Where Kenyan law stands on AI in legal work (November 2026)

Kenya has no statute dedicated to AI. Existing law and professional duties apply, and the court rules written for AI are still drafts.

On LSK guidance: we could not find any published LSK guidelines on AI for advocates. The University of Oxford's AI tracker reported none as at September 2026. In May 2026 the LSK was reported to be forming an ICT and AI Committee to guide adoption and training. Its August 2026 conference was themed on legal practice in the era of AI, and an article in its conference magazine called on the Society to issue AI guidelines. If guidance has appeared since we checked, follow it.

What the draft Practice Directions would require

Reports on the draft describe a disclosure model. As summarised by MMS Advocates in October 2026 and Nairobi Law Monthly in May 2026, advocates would:

• Disclose AI use in pleadings, submissions, affidavits, witness statements and other filings, naming the tool and the kind of help received. Some reports call this a certificate of human verification.
• Check every case, statute and principle from AI-assisted research against authoritative sources.
• Not file a document wholly generated by AI without meaningful human review.
• Correct any error found later and explain to the court how it arose.
• Make reasonable enquiries into a platform's data retention and training practices before using it on client matters.

These are proposals. We could not get the text of the draft directions themselves, so read the final version when it is issued.

The Kenyan cases every firm should know

Kenyan courts have now dealt with irrelevant citations and machine-generated pleadings, and a senior judge has warned about fake ones. We describe them from published reports and legal commentary, so read the rulings themselves on Kenya Law before relying on them.

Why the April 2026 Savla ruling matters most

It goes further than any fake citation case. Here, the litigant said he had reviewed and adopted every document, and the Commission identified no fabricated authority. According to published reports and a Strathmore CIPIT commentary, Chigiti J still held that:

• Generating pleadings through unknown tools or AI gives an unfair advantage (paragraph 81).
• The litigant should have told the court at the start that such tools had been used.
• A judgment flowing from those pleadings could not stand (paragraph 91).

As reported, this litigant was barred from filing machine-generated pleadings in any court unless a Kenyan law comes to provide for AI-assisted drafting. The court also invited the Rules Committee to consider amending the Civil Procedure Rules.

The ruling is contested. A Strathmore CIPIT commentary argues that Order 2 governs content and format, and says nothing about drafting tools. Senior lawyer Ahmednasir Abdullahi criticised it publicly, while Miguna Miguna supported it, according to Nairobi Law Monthly. The draft Practice Directions also lean towards disclosure.

For a firm, the honest position is that this is unsettled. We found no report of an appeal against this ruling or a rule change as of 8 November 2026. Until the rules are final, have a partner decide how AI may touch anything filed in court, and check for new directions before each filing.

What has happened outside Kenya

The same pattern appears worldwide, and courts are putting the responsibility on the lawyer.

• United States, June 2023. In Mata v Avianca, lawyers filed non-existent opinions created by ChatGPT and stood by them when challenged. The court imposed a USD 5,000 penalty, while noting that using AI assistance is not inherently improper.
• South Africa, January 2025. In Mavundla v MEC in the Pietermaritzburg High Court, cited authorities turned out not to exist (one published count is seven of nine). The law firm was ordered to pay the costs of additional court appearances from its own pocket, and the judgment was sent to the Legal Practice Council for investigation.
• England, June 2025. In Ayinde and Al-Haroun [2025] EWHC 1383 (Admin), the Divisional Court warned that free generative AI tools cannot do reliable legal research. In Al-Haroun, 18 of 45 cited authorities did not exist. The solicitor had relied on research done by the client. The court said that managing partners and heads of chambers must take practical steps too.

Damien Charlotin's public database listed 2,266 decisions involving AI hallucinations when last updated on 5 November 2026. Only one was tagged Kenya: Nyanjui, recorded as alleged AI use.

Why AI invents cases

A general AI chatbot predicts likely words. It is not looking the answer up in a law report unless it has been connected to one, so it can produce a perfect-looking citation for a case that never existed.

Paid legal research tools reduce this risk but do not remove it. A 2024 Stanford study found that Lexis+ AI and Ask Practical Law AI gave incorrect information more than 17% of the time on its test queries, and Westlaw AI-Assisted Research more than 34%. Products have changed since, so treat those figures as a warning and test your own tool.

How to verify an authority on Kenya Law

Kenya Law is the free public database of Kenyan judgments and legislation, and it should be your first stop. This check takes a few minutes per authority.

1. Search the neutral citation, which takes the form [year] KEHC number (KLR). Older cases may carry an eKLR citation. If that fails, search the party names.
2. Match every detail: parties, court, judge, date and case number. A real name attached to the wrong citation is still an error.
3. Open the full text. Find the exact paragraph you want to rely on and read it in context.
4. Ask if it fits your point. Nyanjui shows why. Check the facts, the issue and whether the passage is the decision or a passing remark.
5. Check it still stands. Look for later appeals and contrary decisions, using any citation or case-history details shown with the judgment.
6. For statutes, open the Act on Kenya Law, note the version date shown and read the section itself.
7. If it is not on Kenya Law, do not assume it is fake, as some older or unreported decisions may be missing. Get a copy from a law report or the registry. If you cannot get one, do not cite it.
8. Keep a record. Save the judgment and fill in the check sheet below.

Make the rule simple: no authority goes into a draft until a named person has filled in its row.

Where AI helps, and where it does not

Match the task to the risk. The more a mistake would cost a client, an opponent or the court, the less AI should do.

Affidavits deserve special care. An affidavit is a sworn statement, so the deponent's own account has to come first.

Confidentiality and advocate duties

Pasting client material into an AI tool puts it in someone else's system, and your duties do not pause for that. We found no Kenyan decision on whether doing so breaches section 134, so treat it as a serious risk that remains untested.

Privilege. Section 134 of the Evidence Act stops an advocate disclosing client communications, document contents or advice without the client's express consent, subject to limited exceptions. Section 135 applies this to advocates' clerks and servants, so your support staff are covered as well.

Confidentiality is wider than privilege. SOPPEC-7 makes unauthorised disclosure of client confidential information professional misconduct, and the Code says its scope is greater than legal professional privilege (paragraph 109).

Systems count. The Code requires measures and systems for data storage and protection, including electronic communications (paragraph 112). If weak protection lets client information reach third parties, the advocate could face a misconduct finding or a negligence claim.

Competence. SOPPEC-3 requires competent and diligent service, and the Code expects advocates to invest in the technical resources needed to do so (paragraph 51).

Honesty with the court. Section 55 of the Advocates Act makes every advocate an officer of the court. The Code says an advocate must be accurate and must not mislead the court (paragraph 124). Section 60 allows anyone to bring a complaint of professional misconduct.

The Judiciary's own draft sets a useful benchmark. Section 5.2.2 of the draft AI Policy bars judicial officers and staff from putting case files, evidence, witness statements or confidential material into third-party AI tools. Firms can reasonably hold themselves to a standard no lower.

What the tools do with your text (as of November 2026)

These settings change often. Check the current terms before relying on them.

A business plan improves your position, but it is not client consent and does not settle privilege. Your firm still has to decide what may go in. For the most sensitive matters, some firms may prefer an open-source model running on their own machine, with their ICT provider's help.

The Data Protection Act 2019: why it reaches your files

If your firm decides how personal data is used, the Act applies to that processing, and an AI tool adds a new place for the data to go. Key provisions to raise with your data protection lead:

• Section 25: data must be limited to what is necessary and kept accurate, and transfers outside Kenya are restricted.
• Section 41: build protection into how you work, and process only what each purpose needs.
• Section 42: when you use a data processor, have a written contract and make sure it keeps adequate security measures.
• Sections 48 and 49: conditions apply to sending data out of Kenya. Sensitive personal data also needs the data subject's consent and confirmation of safeguards.
• Section 31: high-risk processing needs a data protection impact assessment first.
• Section 35: people have a right not to be subject to decisions based solely on automated processing that significantly affect them, subject to exceptions.
• Section 63: the Data Commissioner can issue fines of up to KES 5 million or 1% of annual turnover, whichever is lower.

The Act's definition of sensitive personal data includes health status, property details, marital status and family details, such as the names of children and spouses. That covers much of what sits in succession, family and land files.

Never paste this into a public AI tool

Use one test: would your client be comfortable seeing this text on a stranger's screen?

• Client names, ID numbers, KRA PINs, phone numbers and addresses
• Instructions, advice, opinions and attendance notes
• Unfiled pleadings, draft affidavits and witness statements
• Medical, family, succession and land details tied to a person
• Deal terms, due diligence material and board papers
• Settlement positions and litigation strategy
• Anything under a court order or confidentiality agreement

Changing a name is often not enough. A rare fact, a named location or a date can identify someone. Here is an illustration:

Too revealing: "Jane W., of Kiambu, wants to contest her late husband's estate because his brother forged a transfer of plot 123 on 4 March."

Safer: "A widow wants to challenge a transfer of land from her late husband's estate. She believes a relative forged it. What issues should an advocate research under Kenyan succession and land law? List questions only."

Five sample prompts you can adapt

We teach four Cs before any fancy method: Clarity (what you want), Context (the background), Constraints (limits and format) and Customisation (who will read it and where). Square brackets show what to replace. Use public or anonymised material only.

1. Understanding a public judgment

I am a junior advocate in Kenya. I will paste a public judgment. Explain it for my own preparation. List the facts, issues, holding and reasoning, with paragraph numbers for each point. Use only the text I paste. Write "not stated" where the judgment is silent. Do not add any other case or statute. Keep it to one page. Judgment: [paste text]

2. A research plan that cannot invent cases

Help me plan research on [legal issue] under Kenyan law. Assume I will search Kenya Law myself. Do not give any case names, citations or section numbers, because I will find those. Give me (1) the sub-questions to answer, (2) search terms to try, (3) the types of law I should check, and (4) facts I need from the client. Format as a checklist.

3. Testing your own argument

Below is an anonymised summary of an argument. Act as opposing counsel in a Kenyan court. List the five strongest objections, starting with the most serious. For each one, say what fact or authority I would need to look for. Do not cite any authority yourself. Argument: [anonymised summary]

4. A plain-language client explanation

Explain [public legal process, for example how a grant of probate works in general] to a client with no legal background. Context: they are anxious and will read this on a phone. Constraints: under 200 words, short sentences, no Latin, no advice on any particular case. Mark every statement of law with [VERIFY] so that I can check it before sending. Customisation: warm and respectful.

5. For law students: test me

I am a Kenyan law student revising [topic]. I will paste my own notes. Ask me ten questions one at a time, wait for my answer, then tell me what I missed, using only my notes. Do not write any part of an assignment. Do not cite any case that is not in my notes. Notes: [paste notes]

Asking for no citations, a "[VERIFY]" marker and "not stated" makes gaps visible. It does not make the output correct, so the check on Kenya Law still applies.

If you are a law student or pupil

Use AI to understand and practise, and do your own legal reasoning. Your institution's rules on AI come first, and they can differ between courses, assessments and examiners. Ask before you assume.

• Never submit AI-written work as your own, and disclose AI help where your institution requires it.
• Never put a case in an assignment that you have not opened and read.
• Learn to read and brief a judgment without help first. You cannot catch an error in a skill you never built.

Pupils have a practical advantage here, because many firms do not yet have a verification routine. Learn the Kenya Law check above and offer to run it. Partners remain responsible for supervision, as the South African and English cases above show. Our guide to the ethical use of AI in academia goes further.

For in-house legal and compliance teams

Your job is usually to give the business a safe route forward, so judge each use by the task, the data, the tool and the controls. Ask:

• What data goes in, and whose is it?
• Where is it processed, and is it retained or used for training?
• Who sees the output, and what decision will someone make because of it?
• Is there a written contract with the vendor, and can we leave with our data?

Set review levels. A meeting agenda needs no sign-off, while staff or customer decisions and new vendors need a deeper look. Put the result in a short policy; our guide to an AI usage policy for Kenyan organisations sets out what to cover.

What goes wrong, and how to avoid it

What a firm's AI training should cover

Good training uses your own kinds of work, on safe documents, and leaves people with habits. Nobody becomes an expert in a few days, so plan for practice afterwards.

A simple 30-day start

• Week 1: A partner owns the issue. Find out which tools people already use, without blame.
• Week 2: Write a one-page rule covering approved tools, the never-paste list, the check sheet and how AI may touch filings.
• Week 3: Run a hands-on session for everyone, including clerks and support staff.
• Week 4: Pick one low-risk task, such as summarising public Bills. Log the time taken, including the checking, and every error caught. Review the rule when the Judiciary's final directions appear.

Training will not settle the legal questions above, and it will not make a poor tool safe.

Questions people ask

Is it illegal for a Kenyan lawyer to use ChatGPT?

No Kenyan statute bans it. The High Court has, however, treated machine-generated pleadings as improper in the April 2026 Savla ruling, and the rules are unsettled. Internal learning and research planning on public material carry much less risk than anything filed in court.

Do I have to tell the court I used AI?

The draft Practice Directions would require disclosure, but they were not final when we checked on 8 November 2026. In Savla, the court criticised the litigant for not disclosing at the start. Check the current directions and take your own advice before filing.

Has the Law Society of Kenya issued AI guidelines?

We could not find any as of 8 November 2026. The LSK was reported to be forming an ICT and AI Committee, and an article in its conference magazine has called for guidance. Its 2017 Code still applies to how you use these tools.

How do I check whether a case is real?

Search the citation and party names on Kenya Law, or find it in a law report. Then read the full decision. Never ask an AI tool to confirm its own citation.

Can I upload a client's contract to AI?

Not to a public tool. Even with a business tool, check the data terms and your section 134 and SOPPEC-7 duties, and consider client consent. If in doubt, leave it out.

Can law students use AI for assignments?

It depends on your institution's rules. Use AI to learn, test yourself and plan, and do not submit its writing as your own. Read every case you cite.

Sources

All last checked 8 November 2026.

Kenyan law and rulings

• Evidence Act, sections 134 and 135, Kenya Law
• Advocates Act, sections 55 and 60, Kenya Law
• Data Protection Act, 2019, Kenya Law
• LSK Code of Standards of Professional Practice and Ethical Conduct, Kenya Gazette, 26 May 2017
• Damien Charlotin database entry, Nyanjui & another v Nyanjui (Land Case Appeal E009 of 2025)
• Nation, "Judge warns lawyers over AI after bizarre LGBTQ case citation in land row"
• Eastleigh Voice, on Savla v Commission on Administrative Justice (20 April 2026)
• Radio Generation, on the same ruling and the bar on further filings (20 April 2026)
• HapaKenya, on the same ruling and the Rules Committee (19 April 2026)

Judiciary, LSK and commentary

• Draft AI Policy for the Judiciary of Kenya (Policy 2.0, hosted by Africa Jurists)
• University of Oxford, "AI in Kenya's Judiciary and Law Enforcement"
• MMS Advocates, "AI in the Kenyan Courts: Draft Practice Directions" (1 October 2026)
• Nairobi Law Monthly, "Judiciary moves to regulate AI in courts" (28 May 2026)
• Nairobi Law Monthly, "High Court ruling sparks debate on AI use" (21 April 2026)
• Techweez, "Judiciary moves to regulate AI use in courtrooms" (15 May 2026)
• The Advocate magazine, LSK Annual Conference Edition 2026
• People Daily, "LSK rolls out AI and legal technology reforms" (8 May 2026)
• Switch TV, "Kenyan judge warns lawyers over fake AI citations in court submissions" (19 May 2025)
• Africa.com, "Why a Kenyan High Court rejected an AI-generated court filing" (12 March 2026)
• Strathmore CIPIT, "When the Drafter is Algorithmic" (23 June 2026)
• Digital Policy Alert, ODPC draft Guidance Note on AI
• Techweez, Senate committee rejects AI Bill (16 October 2026)

Cases and research outside Kenya

• Ayinde and Al-Haroun [2025] EWHC 1383 (Admin), National Archives
• LawSites, on the Mata v Avianca sanctions (June 2023)
• ITLawCo, on Mavundla v MEC [2025] ZAKZPHC 2
• Bizcommunity, "The future of law: 9 citations, 7 fake" (25 September 2026)
• Damien Charlotin, AI Hallucination Cases database
• Stanford HAI, "AI on Trial" (2024)

Tool data terms

• OpenAI, "How your data is used to improve model performance"
• Google, Gemini Apps Privacy Hub
• Anthropic, "Is my data used for model training?"
• Microsoft, Copilot Chat privacy and protections

Where FuKazee fits in

FuKazee trains teams in practical AI for the work in front of them. We assume nobody in the room knows AI yet, we avoid jargon, and we show the free version of every feature so people can keep practising. We do not promote one tool. Since 2022 we have trained more than 3,000 people.

For law firms, in-house legal teams and compliance functions, we can build a session around the checks in this guide: spotting fake authorities, verifying on Kenya Law, keeping client data out of the wrong places and writing better prompts. We teach the skills and the habits. The legal judgement stays with your advocates. See how our AI training in Kenya works.

To discuss training for your firm or legal team, email us at lead@fukazee.com