The Judiciary’s proposed Practice Directions on the Use of Artificial Intelligence in Courts signal an important shift in how artificial intelligence (“AI”) may be approached within Kenya’s justice system. Rather than imposing a blanket prohibition on its use, the draft Practice Directions seek to accommodate AI-assisted legal and judicial work while placing clear responsibility on the human user. The underlying proposition is straightforward: AI may assist the work of an advocate or judicial officer, but it cannot assume their professional or judicial responsibilities.
A significant feature of the proposed framework is the express permission to use AI in judicial proceedings, subject to specified safeguards. For advocates, this extends to research, drafting pleadings and submissions, summarisation, translation and other forms of assistance. The permission, however, is coupled with an obligation of independent professional judgment. An advocate remains fully responsible for the accuracy, completeness and integrity of any document filed in court, irrespective of whether its contents were generated or assisted by AI. The draft therefore treats AI as a tool rather than a substitute for professional judgment.
This principle is particularly pronounced in relation to legal research. The proposed Directions expressly recognise the risk of fictitious, inaccurate or misattributed citations generated by AI and place the burden of verification squarely on the advocate. Every case, statute, regulation, legal principle and secondary source relied upon following AI-assisted research must be independently checked against authoritative sources. Importantly, the obligation is not limited to situations where an advocate knowingly relies on an erroneous citation. The proposed framework makes verification a professional obligation in its own right. Where an error is subsequently discovered, the advocate would also be required to notify the court, correct the document as directed and explain the circumstances in which the error arose.
The draft goes further by introducing a formal disclosure regime. An advocate who uses AI in preparing a pleading, submission, affidavit, witness statement or other filed document would be required to identify the tool used, its version or model where ascertainable, the nature of the assistance obtained and confirm that the resulting content has been independently reviewed and verified. Litigants in person are similarly subject to disclosure requirements, albeit through a separate form. This approach places transparency alongside accuracy: the court is not simply concerned with what AI produced, but with whether its use can be identified and accounted for.
Another significant area is AI-generated evidence. The proposed Directions contemplate that AI-generated text, images, audio, video and other content may be tendered as evidence, but subject to the applicable evidentiary requirements and additional safeguards. A party intending to rely on such material would have to disclose, among other things, the AI tool used, the date and manner of generation, the prompt used, the purpose for which the material is tendered and any subsequent modifications. The party would bear the burden of establishing authenticity, accuracy and relevance. The treatment of deepfakes is particularly cautious: AI-generated material manipulating the appearance, voice or conduct of real persons would require expert-supported consideration before admission.
The proposed framework also recognises that the risks associated with AI extend beyond accuracy. Its provisions on data protection and confidentiality require compliance with the Data Protection Act and place particular emphasis on information subject to legal professional privilege, confidentiality obligations and sensitive personal data. Advocates would be expected to make reasonable enquiries into an AI platform’s data retention and training practices before using it in connection with client matters. This is a practical reminder that the question is not merely whether an AI tool can perform a task, but whether it can be used to perform that task without compromising duties owed to clients or the court.
For Judges and Judicial officers, the proposed safeguards are more stringent. AI may assist with legal research, case organisation, translation and administrative functions, but judicial decision-making, reasoning and discretion cannot be delegated to an AI tool. Judges would also be prohibited from using unapproved platforms to process confidential case information and would remain responsible for independently verifying AI-assisted content in judgments, rulings and orders. The draft consequently preserves human judicial authority while recognising limited technological assistance.
For legal practitioners, the proposed framework does not make AI unusable in legal practice; it makes uncritical use of AI professionally untenable. If adopted, the Directions would require advocates to treat AI-assisted work in much the same way they would any other professional work product: verify it, exercise independent judgment, protect confidential information, disclose its use where required, and take responsibility for the final product.
