Lately, AI has been finding its way into Kenyan courtrooms, and not always gracefully. Lawyers are using it to draft filings, research cases and build arguments, often quietly, and sometimes with consequences that end up before a judge.
When the Milimani High Court struck out an AI-drafted legal filing in March 2026, as reported by the East African, it sent a message to the legal profession. A generative AI tool had completely assembled the application, which contained inaccuracies, including citations to non-existent cases.
Two months after the judge had the filing thrown out, the Judiciary has responded with something more structural, releasing a Draft Judiciary Artificial Intelligence Policy that attempts to bring order to a profession increasingly tempted by, and tripped up by, AI tools.
The new draft serves as a constitutional blueprint, grounded in Article 159 of the Constitution, which requires that judicial authority be exercised in ways that promote justice for all.
The policy operates within that mandate, drawing clear lines between where AI may assist and where human judgement remains non-negotiable.
The policy calls the human-in-the-loop requirement, where judicial officers must remain the decision-makers. AI tools may assist with research, case management, scheduling, and transcription, but they cannot replace the bench.
Any AI-assisted filing must be accompanied by a certificate of human verification, a direct response to the hallucination problem that felled the Milimani application.
The certificate places accountability back on the advocate, making clear that the lawyer, not the algorithm, is responsible for what lands before a judge.
The policy sorts AI applications into three risk tiers, borrowing the same logic as Kenya’s broader Artificial Intelligence Bill, 2026.
Tools that carry the heaviest consequences, such as those used for bail assessment or predictive sentencing, require mandatory human oversight and regular audits.
Legal research platforms sit in the middle tier, where lawyers must verify outputs and disclose their use of AI.
At the lower end, scheduling and transcription tools face only general monitoring. The tiered structure is deliberate so that it avoids the blunt instrument of an outright prohibition while ensuring the most consequential decisions carry the heaviest scrutiny.
AI systems used in court proceedings will handle sensitive personal information about litigants. The policy requires that this data be managed in line with the Data Protection Act, 2019.
The policy also raises the risk of bias, warning that AI left unchecked could produce unfair outcomes for people who are already at a disadvantage in the justice system. It goes a step further, taking on the question of who gets left behind entirely.
Efficiency gains from AI are of limited value if they widen the gap between those with digital literacy and those without. The Judiciary proposes that ICT help desks at Huduma Centers be positioned to ensure that technology expands access rather than creating new barriers to it.
The policy also aligns with UNESCO’s 2025 guidelines for AI use in courts and tribunals, which also insist on transparency and the non-delegation of judicial decisions to machines.
Kenya’s draft goes further in one respect: it proposes linking the Judiciary, the police, the Office of the Director of Public Prosecutions, and the prisons service into a single connected digital system.
The draft is now in the public participation phase, and what it ultimately achieves will depend on implementation.
