Looking for feedback on how to regulate deepfakes, chatbots and AI agents
The federal government has launched a public consultation seeking Canadians' views on how to make artificial intelligence systems more transparent, opening a two-month window for input that will help shape legislation and policy in one of the fastest-moving areas of technology regulation.
The consultation will run until Sept. 23, 2026. It centres on five areas where the government says existing rules and market practices have not kept pace with the technology:
- identifying AI-generated content
- disclosing when Canadians are interacting with an AI system
- publishing meaningful information about how AI systems work and were built
- reporting serious AI-related incidents
- tracking the growing use of AI agents that take automated actions on behalf of users.
"AI adoption moves at the speed of trust," said Evan Solomon, Minister of Artificial Intelligence and Digital Innovation.
"Canadians need to know when they are interacting with AI systems, when content has been generated or altered by AI, and what these systems can and cannot do."
Why this matters to HR
The federal government's consultation may read like a technology policy story, but its implications land squarely on HR. Ontario already requires employers to disclose when AI is used to screen job applicants — and this consultation signals that similar obligations are coming federally, covering hiring, performance management, and other consequential decisions about workers.
The paper puts deployers of AI systems, not just developers, on the hook for transparency and accountability, which means HR teams that buy and use third-party AI tools own the compliance problem. It also sets a high bar for what disclosure means: telling someone AI was involved may not be enough — employers may need to explain what the system does, what personal data it uses, and what human oversight is in place.
And if mandatory AI incident reporting follows, HR departments using AI in decisions that affect workers may face new documentation and reporting obligations as well.
Deepfakes and the cost of synthetic content
A substantial portion of the discussion paper, Enhancing trust in artificial intelligence through increased transparency, focuses on AI-generated content and the harms that can follow when Canadians cannot tell what is real.
The government points to the Canadian Anti-Fraud Centre's warnings that scammers are increasingly using generative AI tools to create deepfakes of prominent individuals to promote fraudulent investments. The discussion paper references a 2023 case in which an Ottawa couple lost a significant sum after acting on the advice of a deepfake video promoting a fraudulent investment opportunity.
A KPMG survey of large Canadian businesses, cited in the paper, found that nearly three-quarters of respondents reported losing between one and five per cent of their annual profits to AI-powered fraud over the preceding year.
Canada currently has no requirement for developers or deployers of generative AI systems to disclose when content is AI-generated, the paper notes. Bill C-34, the Safe Social Media Act currently before Parliament, would require regulated social media platforms to label synthetic content that could be mistaken for authentic recordings — but that obligation falls only on platforms where the content is shared, not on the AI companies that generate it.
The paper reviews technical approaches such as invisible watermarks and provenance metadata, noting that the Coalition for Content Provenance and Authenticity — whose members include Amazon, Google, Meta, Microsoft and OpenAI — has developed a technical specification for attaching certified origin information to digital content.
However, the paper cautions that determined bad actors can strip or evade these signals, and that technical approaches "struggle to keep pace with rapid technological advances."
When the chatbot doesn't say it's a chatbot
The consultation also asks when Canadians should be told they are talking to a machine rather than a person — and what that disclosure should look like.
The discussion paper illustrates the stakes with a case reported by CBC News in June 2026: a Toronto man who tried to sell his car back to a dealership was offered a buyback price by a text-based contact named "Quinn," who set a meeting time and offered to consider a counteroffer. Only after a sales consultant called to revoke the offer did the man learn that Quinn was an AI chatbot that had made the offer in error.
The dealership ultimately honoured the original price after the man brought his story to the media.
"At no point in the man's interaction with Quinn did it indicate that it was not a real human," the discussion paper states.
Several jurisdictions have already moved on this issue, the paper notes. The EU AI Act requires that AI systems intended to interact directly with users be designed so that users know they are talking to an AI. California and New York require operators of companion chatbots to notify users they are communicating with a machine.
AI hallucinations in court
On the question of what information Canadians should have about AI systems' capabilities and limitations, the government points to a study cited in the discussion paper that reviewed Canadian legal decisions between January 2024 and March 2026 and identified 132 cases in which at least one party cited a fictitious case as legal authority — with courts or tribunals finding in 96 of those cases that the fabricated citations had been generated using AI.
"This illustrates how AI systems can produce highly persuasive and authoritative-sounding outputs that are factually incorrect," the paper says, "especially when systems are deployed outside of their intended context or where their limitations and risks are poorly understood."
The government is considering whether to require AI developers to publish standardized documentation about their systems — similar to voluntary "model cards" that some major developers already publish — and whether training data, including the use of copyright-protected material, should be disclosed to users, researchers, and rightsholders.
The problem of AI agents
Perhaps the fastest-evolving area covered by the consultation is AI agents: systems that can autonomously take sequences of actions on a user's behalf, such as booking meetings, filling out forms, completing online purchases, or managing email.
The discussion paper notes that agentic capabilities have already been integrated into widely used consumer products. OpenAI's Operator, integrated into ChatGPT, can browse the web, fill forms, and complete booking-style workflows. Google's Gemini can read email and calendar context, propose meeting times, and create calendar invites. Amazon's Alexa for Shopping can add items to carts and complete purchases on a user's behalf.
The paper highlights a consumer scenario in which a user instructs an AI agent to automatically repurchase a household item once a month without confirming each transaction, and the agent misinterprets the instruction and orders 12 packages instead of one. "Responsibility for the malfunction is not straightforward to assign between the consumer, the business operating the agent, and the original developer of the agent," the paper says.
The government also notes that some businesses have taken the position that users bear responsibility when agents misbehave — a reference to a case reported in April 2026 in which a major retailer warned that customers would be liable for purchases made in error by its AI shopping agent.
Canada currently has no regulatory framework specifically governing agentic AI systems. The discussion paper says the government is considering disclosure requirements around what actions agents are empowered to take, how human oversight is maintained, and how to track accountability when agents interact with one another in multi-agent systems.
Next steps for AI in Canada
The consultation supports “AI for All,” Canada's national AI strategy launched on June 4, 2026, which identifies protecting Canadians and safeguarding democracy as one of its six core pillars.
The government is accepting responses through an anonymous online survey or by email at [email protected] until Sept. 23, 2026. Submissions will be treated as public documents and may be posted online.
Following the consultation period, the government has committed to publishing a What We Heard report summarizing the input received.
The consultation is separate from, but connected to, several pieces of legislation already moving through Parliament:
- Bill C-36, the Protecting Privacy and Consumer Data Act
- Bill C-34, the Safe Social Media Act
- Bill C-16, the Protecting Victims Act targeting non-consensual deepfakes
- Bill C-25, the Strong and Free Elections Act targeting AI-enabled electoral misinformation.
