The federal government's is seeking to understand where transparency about AI systems and AI-generated content matters most to Canadians, and where further action might be warranted.Bhawika Chhabra/Reuters
Ottawa is consulting Canadians on what action should be taken to deal with AI agents that mislead people with false information, or act against their interests such as shopping for more products than a customer ordered.
The federal government has launched a formal consultation to gauge Canadians’ views on how to deal with the increasing use of AI technology, including asking about labelling AI-generated content, such as music, written reports and videos, with watermarks. The consultation follows the launch by Evan Solomon, Minister of Artificial Intelligence and Digital Innovation, of the federal AI strategy in June.
The consultation is seeking views from Canadians, including tech experts, on what action should be taken to make the use of AI transparent and to address mistakes resulting from its use. The consultation, which includes a survey, will run until Sept. 23 and will inform the government’s next steps.
The survey is accompanied by a detailed discussion paper, published on Thursday, which sets out five areas for consideration.
One question the discussion paper poses is who should be responsible if an AI chatbot gives a person the wrong information.
The paper says AI is being increasingly used by Canadians who “may now encounter AI agents that they can direct to do their shopping, send their emails, or book their meetings.”
But it adds “the growing adoption of AI agents challenges our understanding of liability and heightens AI-related risks.”
We snagged a campsite with an AI bot. Here’s what it means for impossible campground bookings
The paper includes an actual example of a man in Toronto in May who submitted an online inquiry to a car dealership about selling his car. A text communication with him said it would offer to buy back the vehicle for $27,163. But he then received a call from a sales consultant who explained that he had been communicating with an AI chatbot who made the offer in error. The agent, who identified itself as Quinn, had not indicated to the car owner that it was not human. The most the dealership would pay would be $20,000, the sales consultant said. Later it agreed to pay the amount the AI had quoted.
The discussion paper says that several countries and jurisdictions, including the EU and a number of states in the U.S. have put in place requirements to ensure that users know when they are interacting with an AI system.
But it added that AI agents are now interacting with other AI agents, making it more difficult to identify which agent did what, when and why.
“Multi-agent interactions introduce reliability challenges due to miscoordination or conflicts between agents, as they have both shared and individual goals,” the discussion paper says. “Additionally as AI agents and multi-agent systems become more powerful, the consequences of malfunctions, such as executing unauthorized actions, can become more significant for Canadians and society at large.”
It adds that AI agents can increasingly be used for malicious purposes, including cyberattacks and identity theft.
“Shopping agents can be tricked by malicious actors to act against the shopper’s best interests, potentially buying products that are too expensive or not what the user desires,” the discussion paper says.
While AI agents employed to buy goods can make Canadians’ lives easier by automatically buying weekly groceries from an e-commerce site, “they can also malfunction, with potentially serious financial consequences.”
In one case study in the paper, a consumer gives an agent access to their credit card and shopping account. But the AI orders twelve packages of a product, rather than a single package of 12, spending far more than the consumer intended.
“Responsibility for the malfunction is not straightforward to assign between the consumer, the business operating the agent, and the original developer of the agent, and the consumer’s avenues for redress may be unclear,” the paper said. Some businesses have decided their users are responsible when AI agents don’t behave as directed, it added.
The discussion paper warns that AI systems are able to produce persuasive and authoritative-sounding outputs that may be factually incorrect.
It cites a review of decisions on the Canadian legal database between January 2024 and March 2026 identified 132 Canadian decisions issued by 44 different courts and tribunals in which at least one party cited a fictitious case as legal authority. The consultation said the court or tribunal found or inferred that the fictitious cases had been generated using AI in 96 of those 132 decisions.
Generative AI enables the creation and dissemination of deceptive or misleading content with serious and even life-threatening consequences, the paper warns.
“Realistic AI-generated content has already appeared in financial scams targeting Canadians and during events like wildfire emergencies,” the paper said.
The government is seeking to understand where transparency about AI systems and AI-generated content matters most to Canadians, and where further action might be warranted. It said in the discussion paper that it wants feedback on whether legislation, voluntary codes and other policy instruments should be used to enhance transparency.
This consultation follows the recent introduction of the Safe Social Media bill, which if enacted, would make operators of regulated social media companies label synthetic content on their platforms. This could include identifying images, video, audio, and text created or modified by AI.
The consultation seeks views on whether invisible, or visible watermarks, disclaimers, or metadata citing the provenance of the content would best help determine which content is AI generated.
Opinion: Canada has entered a new era of AI in political ads. We will survive
Many generative AI developers now mark content with watermarks. But the discussion paper warns that, while approaches for marking audio, images and video are quite well developed, watermarks applied to text “remain brittle and easy to remove.”
“This consultation will help ensure Canadians have a voice in shaping practical and effective transparency measures,” Mr. Solomon said in a statement.