The Walrus Finds Ottawa Has No Binding Rules for Generative AI as Public Servants Turn to Public Chatbots
Canada’s federal government is encouraging public servants to use artificial intelligence without a binding framework for generative tools, according to an investigation by Colin Horgan published by The Walrus on November 10. The piece opens with Kémy Adé, a McMaster University postdoctoral fellow whose permanent-residence application was refused over job duties she never claimed, such as wiring control circuits. As the Toronto Star first reported, the refusal letter disclosed that generative AI had been used in processing. Immigration, Refugees and Citizenship Canada says a human always makes the final decision.
The only binding federal AI policy remains the 2019 Directive on Automated Decision-Making, which requires a published risk assessment before automated systems that help decide benefits, permits or entry to Canada go live. The Walrus notes that it was written for rule-based systems, not for the chatbots employees reach for on their own. Guidance from the Treasury Board and a voluntary code from Innovation, Science and Economic Development Canada followed in 2023, but neither is mandatory. A KPMG survey of public-sector workers, including provincial and municipal staff, found that nearly half use AI frequently, that half of those rely on publicly available tools, and that 60 per cent had received no data literacy training.
What it means in practice
Shared Services Canada has piloted CANchat, a secure interface that gives employees access to several models for unclassified work, but sources told The Walrus the rollout has barely moved, which can push staff toward workarounds on personal devices. IRCC would not name the public tools its officials use and confirmed that training is not mandatory, while Employment and Social Development Canada requires its staff to complete generative AI training by March 2027. The federal AI Register now lists 412 declared systems, and Western University’s TAG Register had independently documented more than 300 entries.
The findings sit alongside a wider public-sector push, from Quebec testing Cohere in its public service to the federal consultation on AI transparency. What to watch is whether the Treasury Board extends binding rules to generative tools, whether training becomes mandatory across departments, and whether the federal auditor general reviews how staff actually use AI, as Ontario’s auditor general did for that province’s public service.