Quebec · Law 25 (P-39.1)

Quebec’s judiciary has also written down what it expects of AI

On 31 August 2026 the Chief Justices of Quebec signed Lignes directrices sur l’utilisation de l’intelligence artificielle générative par les juges, announced on 4 September 2026 and signed by Geneviève Cotnam, Marie-Anne Paquette, Henri Richard and Nathalie Duchesne. They are recommendations, not a practice direction, and the self-assessment annex is expressly of no normative value. They govern judges, not advocates, and they sit alongside Law 25 rather than inside it.

Three provisions are worth reading even if you never appear in Quebec. The core rule: no judge may delegate the decision-making function to a generative AI system, because «Juger est un acte exclusivement humain», judging is an exclusively human act. On responsibility: the judge remains responsible for the accuracy of references, citations and authorities, and «Aucune erreur ne peut être imputée à l’outil utilisé», no error can be attributed to the tool used. And section 6 requires that documents supplied by the parties or by third parties are not passed to a tool without prior checking of their content, provenance and format, because they may contain hidden information or instructions capable of unduly influencing the tool or compromising its reliability. That is the first judicial instrument in this record to name hidden-instruction attacks.

The self-assessment annex asks the judge to confirm, among other boxes, «j’ai conservé une trace de ma démarche préalable», that a record of the prior process has been kept.

Working translations, made from the French text read at source on 11 September 2026. No official English version exists. The guidelines, ten pages · the evidence duties timeline