Introduction
The ruling in Commission on Administrative Justice v Savla [2026] KEHC 4891 (KLR) is one of Kenya’s first direct judicial statements on artificial intelligence in court pleadings. Its concern is understandable. Courts must guard against false authorities, unreliable factual claims, and documents filed without human responsibility. However, the ratio goes too far when it treats AI assisted drafting as unlawful merely because the Civil Procedure Rules do not expressly provide for it.
The better approach is not to prohibit the tool. It is to regulate its misuse. A pleading should stand or fall on compliance, truthfulness, attribution, verification and prejudice, not on whether a digital tool helped the litigant draft it.
What AI Means in Litigation
AI should be defined with care. In litigation, it includes systems that perform tasks usually associated with human reasoning or language processing, including legal research, drafting, summarising, translation, transcription, classification, prediction and document review. Generative AI is narrower. It refers to tools that generate text, images, audio or other content from prompts.
That distinction matters. Courts already use technology that may contain AI features. Virtual hearings on Microsoft Teams, live transcription, captions, speaker recognition and intelligent recap tools all involve automated language processing to some degree. A court cannot sensibly condemn “AI” in absolute terms while using AI enabled infrastructure to run hearings, produce transcripts or improve accessibility. The proper question is not whether AI was used. It is what type of AI was used, for what purpose, with what verification, and with what effect on fairness.
Access to Justice and the Self Represented Litigant
The ruling assumes that AI gives the user an unfair advantage. That assumption is too broad. AI can create risk, but it can also reduce exclusion. For self represented litigants, AI may help them understand procedure, organise facts, improve grammar, identify issues and present grievances in a form the court can follow.
This matters where legal services remain expensive and procedure is technical. A constitutional commission, public body or commercial entity has lawyers, precedent banks and institutional memory. A self represented litigant often has none. Treating responsible AI assistance as an unfair advantage may deepen the inequality Article 48 seeks to reduce. Access to justice should support tools that help litigants participate, subject to safeguards against falsehood and abuse.
The Civil Procedure Rules Were Read Too Narrowly
The Court relied on Order 2 of the Civil Procedure Rules. Order 2 regulates the form and content of pleadings. It requires material facts, numbered paragraphs, brevity and clarity. It does not regulate the private tools used before a pleading is filed.
A purposive interpretation should ask whether the pleading performs its function. Does it notify the opposing party of the case to be met? Does it define the issues? Does it assist the court to deal with the dispute fairly? If the answer is yes, the use of a computer, precedent bank, grammar tool, research database, transcription software or generative AI should not invalidate it.
Article 159 also matters. Procedure must serve justice. It should not become a basis for excluding a litigant because the rule is silent on a modern drafting tool. Silence in the Rules is not prohibition.
Comparative Jurisprudence: Misuse, Not Use
The comparative cases relied on in the discourse around AI do not support a blanket ban. They condemn misuse.
In Mata v Avianca Inc, the United States District Court for the Southern District of New York sanctioned lawyers because they filed non existent cases generated by ChatGPT. The vice was not AI use. It was the filing of false authorities and the failure to verify them.
The same distinction appears in Park v Kim, 91 F 4th 610 (2d Cir 2024), where counsel cited a non existent authority and admitted she had not confirmed it. Again, the problem was not the tool. It was professional failure.
The English decision in Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank QPSC [2025] EWHC 1383 takes the same practical position. The Court accepted that AI may be useful in litigation, but warned that lawyers must verify all material before placing it before a court.
Singapore is even clearer. In Tan Hai Peng Micheal and another (as the executors of the estate of Tan Thuan Teck, deceased) v Tan Cheong Joo and another and other matters [2026] SGHC 49, the High Court relied on the Australian decision in JNE24 v Minister for Immigration and Citizenship [2025] and stated that “it is not the initial reliance on AI that constitutes the vice”. The impropriety lies in placing false authorities or evidence before the Court.
Guidelines Are Better Than Prohibition
Kenya should follow jurisdictions that regulate AI use directly. The Federal Court of Canada requires a declaration where litigation materials contain AI generated content, but clarifies that the declaration is only to notify the Court and parties. It is not an automatic adverse inference. The user remains responsible for accuracy and truthfulness.
Singapore’s Guide on the Use of Generative Artificial Intelligence Tools by Court Users, which applied from 1st October 2024, takes a similar route. It allows use of generative AI but requires court users to verify accuracy, ensure appropriateness, and remain responsible for what is filed.
These models offer Kenya a practical path. The Judiciary and Rules Committee can issue a practice direction requiring disclosure where generative AI materially contributes to filed documents, independent verification of authorities, protection of confidential information, and certification that a human litigant or advocate adopts the final document.
The Role of the LSK
The Law Society of Kenya should also act. Its mandate includes maintaining standards of professional practice, ethical conduct and competence in the provision of legal services. That mandate is wide enough to support AI guidance for advocates.
The LSK need not wait for Parliament. It can issue practice guidelines on competence, confidentiality, client consent, verification of authorities, use of paid or public AI tools, data protection, billing transparency and supervision of pupils or staff using AI. It should also distinguish between advocates and self represented litigants. The LSK regulates advocates, not the general public. Its guidelines should bind members of the profession while informing broader court practice.
Conclusion
The ratio in Commission on Administrative Justice v Savla is difficult to defend because it treats AI assisted drafting as unlawful without proof of false authority, false evidence, misleading citation or prejudice. It mistakes silence in the Civil Procedure Rules for prohibition and risks undermining access to justice.
The correct position is narrower and defensible. AI may assist. Humans must verify, adopt and remain accountable. Courts should punish falsehood, non compliance and abuse, not the responsible use of technology. Kenya should move toward disclosure, verification and professional guidance, not blanket judicial suspicion of AI.
