Source: The Globe and Mail

Ottawa is consulting Canadians on responses to AI agents that can mislead people with false information or act against users’ interests, such as shopping bots that buy more than intended. The federal discussion paper and accompanying survey, launched following Minister Evan Solomon’s June federal AI strategy, seek public and expert views and will run until Sept. 23 to inform next steps on transparency and remedies. The paper poses questions about allocation of responsibility when an AI chatbot supplies incorrect information and gives a May Toronto example where a dealership chatbot named “Quinn” quoted $27,163 for a car, later said to be an error by a human salesperson, with the dealer ultimately paying the AI-quoted amount.

The paper highlights risks from multi-agent interactions, noting miscoordination and conflicts between agents can lead to unauthorized actions with significant consequences. It warns shopping agents can be manipulated to purchase goods not wanted or at excessive cost and gives a case study where an agent with access to a consumer’s credit-card and shopping account ordered twelve packages rather than a single pack of twelve, causing greater expense. The discussion paper states responsibility between consumer, business operating the agent, and original developer may be unclear and that some businesses have treated users as responsible. It also flags persuasive but incorrect AI outputs and cites a review finding 132 Canadian court and tribunal decisions (Jan. 2024–Mar. 2026) where at least one party cited a fictitious case, with courts or tribunals finding or inferring AI-generation in 96 of those decisions. The consultation asks whether legislation, voluntary codes, visible or invisible watermarks, disclaimers, or provenance metadata would best increase transparency, noting watermarking techniques for audio, images and video are more developed than brittle text-watermarks.