Project co-director Maura Grossman said a Canadian team’s court evidence tool did not produce useful agreement among its algorithms in two real-world attempts, complicating efforts to assess potentially AI-generated evidence in court.
Grossman told the Canadian Press that the results from those attempts were less encouraging than the team’s laboratory work. The prototype brings together several detection methods, she said, but their answers were too inconsistent in the reported attempts to guide users. Grossman called the challenge a “wicked technical problem.”
The project aims to create a free, open-source tool for justice-system users to flag evidence that may have been generated or manipulated with AI. Its description says the system is meant to support human judgment. The University of Waterloo said intended users include people representing themselves in court and court officers.
The Canadian Institute for Advanced Research (CIFAR) announced the initiative in November 2025 through the Canadian Artificial Intelligence Safety Institute (CAISI) Research Program. The court-focused effort is one of two AI Safety Solution Networks, each awarded $700,000 over two years. The announcement did not identify the currency. Grossman is co-directing the network with University of Toronto professor Ebrahim Bagheri.
The University of Waterloo said expert review can be expensive and delay trials, a burden the project is intended to help address for litigants and court officers. Grossman also said digital files can change before they reach court: images shared online may be compressed or lose metadata, and transfers between people can alter data. By the time a dispute is heard, she said, the original file may no longer be available.
Grossman said the team plans to seek input from law enforcement and policymakers. The effort adds to discussion about human responsibility in judicial decisions, including Singapore Chief Justice Sundaresh Menon’s call for human responsibility in courts.
