Quebec Judiciary: AI Guidelines Criticized as Deterrent, Not Guide

• Quebec's new judicial AI guidelines, released September 4, are criticized for being overly restrictive and discouraging AI use among judges.
• The guidelines, signed by Chief Justice Geneviève Cotnam and other top judicial officials, prohibit AI for core judicial functions like legal reasoning and fact analysis.
• Lawyer Dmitry Shniger argues the framework primarily lists prohibitions, offering little practical guidance and making compliant AI use nearly impossible.
• Restrictions extend to confidential data, party documents, and judges' notes, due to concerns about data security, hidden instructions, and the lack of secure, authorized AI tools.
• The judiciary justifies its cautious stance by citing the opacity of commercial AI algorithms and the absence of institutionally deployed secure AI tools.

As Shniger sees it, the document primarily functions as a catalogue of prohibitions, effectively discouraging judges from engaging with the technology rather than providing a framework for its responsible integration.

Quebec's judiciary has unveiled new guidelines governing the use of generative artificial intelligence by judges, a framework that a prominent legal expert suggests is designed more to deter than to guide. Released on September 4, the document bears the signatures of Chief Justice Geneviève Cotnam of Quebec, Superior Court Chief Justice Marie-Anne Paquette, Court of Québec Chief Judge Henri Richard, and Chief Municipal Judge Nathalie Duchesne, indicating a unified stance from the province's top judicial authorities.

The introduction of these guidelines follows a period of heightened scrutiny regarding AI's potential impact on legal proceedings. Months prior to their publication, a report by La Presse highlighted a Quebec judgment that allegedly cited non-existent case law, raising concerns that generative AI might have been employed in its preparation. This incident underscored the urgent need for clear policies, even as an April survey by Canadian Lawyer indicated that the Quebec Court of Appeal and the Quebec Superior Court were already among the Canadian courts with the most advanced AI policies in place.

At the heart of the new framework is the fundamental principle that the act of judging is an exclusively human endeavor, and therefore, the critical function of decision-making cannot be delegated to artificial intelligence. While this core tenet is broadly accepted, Toronto lawyer Dmitry Shniger, whose work focuses on human-centered legal design, argues that the guidelines' implementation of this principle is deeply flawed, primarily serving as a list of prohibitions rather than constructive instruction.

Dmitry Shniger contends that the Quebec judicial AI guidelines are overwhelmingly prescriptive in what judges *cannot* do, leaving little room for practical, compliant AI use. He agrees with the explicit prohibitions on using AI for core judicial tasks such as legal reasoning, analyzing factual evidence, weighing credibility, determining outcomes, or drafting any portion of a judgment's rationale. However, Shniger expresses significant disappointment with the document's overall form, arguing that effective guidance should delineate both permissible and impermissible applications of technology.

As Shniger sees it, the document primarily functions as a catalogue of prohibitions, effectively discouraging judges from engaging with the technology rather than providing a framework for its responsible integration. This approach, he suggests, leaves judges with only two viable options: either abstain from AI use whenever there is any ambiguity, a course of action expressly recommended by the guidelines, or conclude that using AI in a manner compliant with the rules is simply not feasible. The lawyer highlights that the guidelines' narrow definition of permissible AI use—limited to producing strictly descriptive summaries that neither interpret nor conclude—is unworkable. Such summaries, which merely restate what was said, in what order, and by whom, and are verifiable against the original source, fail to leverage AI's inherent strengths, particularly its capacity to identify inconsistencies.

The Quebec guidelines impose significant restrictions on the type of information judges can feed into AI systems, particularly concerning confidentiality. While not an outright ban, the framework effectively discourages the input of confidential data, noting that very little material related to judicial decisions is not confidential. It explicitly advises that any information whose confidentiality is in doubt should be excluded, a provision that Shniger points out renders much of a judge's working material unusable with AI.

Further complicating matters, judges are prohibited from inputting documents provided by parties or third parties into AI tools. This restriction stems from concerns that such documents might contain hidden instructions designed to skew the AI's output, unless their content has been thoroughly verified. However, the guidelines offer no definition or criteria for what constitutes such appropriate verification, leaving a critical ambiguity. Additionally, judges are explicitly forbidden from feeding their personal notes on ongoing files into AI systems. Shniger questions the practical utility of AI under such stringent conditions, asking what remains for a judge to act upon once notes, documents, and confidential information are removed from consideration, concluding that the guidelines are confusing except for their directives on corrections and mechanical errors.

The Quebec judiciary's highly cautious approach to artificial intelligence is rooted in specific practical and technical considerations, as outlined in the guidelines' preamble. A primary justification for the restrictive framework is the current absence of any authorized or secure AI tools that have been deployed within the four courts involved. This lack of institutional-grade, vetted technology means that judges would, by default, be reliant on commercial AI tools.

The guidelines express significant reservations about these commercially available platforms, noting that they operate on algorithms whose internal workings remain largely opaque. This lack of transparency regarding how commercial AI processes information and generates outputs contributes to the judiciary's reluctance to endorse their widespread use in sensitive legal contexts. Despite the current limitations, the courts have committed to providing periodic training and institutional support as an accompaniment to the guidelines, signaling an acknowledgment of AI's evolving role while maintaining a conservative stance on its immediate application in judicial decision-making.

The highly restrictive nature of Quebec's judicial AI guidelines means lawyers practicing in the province should exercise extreme caution when using AI for any substantive legal work that might be presented to the courts, as judges are being discouraged from its use and may view AI-assisted work with skepticism. This also signals a broader trend of judicial bodies adopting a conservative stance on AI, which could inform compliance strategies in other jurisdictions.

Source: Original reporting via legal industry commentary.