The federal government has opened a formal public consultation on artificial intelligence transparency, asking Canadians to weigh in on AI agents that mislead people and on how AI-generated content should be labelled. The consultation, hosted on the Innovation, Science and Economic Development Canada (ISED) website under the title "Have your say: Advancing AI transparency in Canada," includes a survey and runs until September 23. (The Globe and Mail; ISED)

A discussion paper published alongside the consultation sets out five areas for consideration. The paper tackles questions that sit at the intersection of consumer protection, civil liability (the legal responsibility to pay for harm caused), and platform regulation — questions the government acknowledges existing laws were not built to answer.

One central question is who should be held responsible when an AI chatbot gives someone wrong information. The paper describes AI agents being used by Canadians to shop, send emails, and book meetings, and warns that these tools challenge existing understandings of who is legally on the hook when things go wrong. The concern is concrete: the paper cites a real case in which a Toronto man received an erroneous $27,163 vehicle buyback offer from a car dealership's AI chatbot, named Quinn, which did not identify itself as non-human. The case shows the kind of consumer-protection gap the consultation is designed to address. (The Globe and Mail)

The paper also raises a newer concern: AI agents are now interacting with other AI agents in what are called multi-agent systems. This goes beyond the familiar scenario of one person talking to one chatbot. When several AI agents work together — say, one bot negotiating a purchase with another bot representing a seller — it creates reliability challenges from miscoordination or conflicts between the agents. That raises questions about cascading failures, how to assign fault, and what disclosure rules should apply when no human is directly involved in a transaction.

On the international front, the paper notes that the European Union and several U.S. states have already put in place requirements to ensure users know when they are interacting with an AI system. Canada has not yet enacted comparable disclosure mandates at the federal level. The consultation appears designed to inform whether and how Ottawa might close that gap, though the government is framing this as an open question rather than a predetermined outcome.

The consultation follows the launch of Canada's federal AI strategy in June. (The Globe and Mail; ISED) Together, the strategy and the consultation signal that the federal government is moving from broad AI policy commitments toward specific transparency and accountability measures. The five areas in the discussion paper, and the responses the consultation gathers, will likely shape whatever laws or regulations follow.

The broader context here is one of jurisdictional and timing pressure. The EU's AI Act and various U.S. state-level measures have established disclosure and transparency rules that Canadian companies operating across borders may already be following in practice. A federal consultation that runs until late September means any resulting policy would arrive well after the House of Commons resumes sitting, and likely after the AI agent market has evolved further. The inclusion of multi-agent system risks in the discussion paper acknowledges that the policy target is moving. Whether a consultation-to-legislation timeline can keep pace with how quickly these systems are being deployed is a question the September 23 deadline does not answer.

For those working in the field, the key detail is the scope. This is not a consultation about general AI safety or research funding. It is specifically about transparency: when and how users must be told they are dealing with an AI system, who bears liability when that system makes an error, and what rules should govern agents acting on their own on behalf of users. The five areas in the discussion paper, the survey, and the September 23 closing date define the narrow window in which affected parties can shape the federal approach.