Canada has revived its online harms agenda, and this time it comes with teeth for artificial intelligence. On June 10, 2026, Minister of Canadian Identity and Culture Marc Miller tabled Bill C-34, the Safe Social Media Act, which enacts two new statutes: the Digital Safety Act and the Digital Safety Commission of Canada Act. The bill is now at second reading in the House of Commons, having cleared first reading on the same day.
The headline provision for this publication's beat is the Duty to Act Responsibly — a statutory obligation, tailored specifically to "regulated chatbot services," that requires operators to mitigate the risk of harmful content, implement crisis-intervention measures, and block manipulative engagement techniques. Where a user expresses suicidal ideation or intent to self-harm, the bill requires the chatbot to "immediately interrupt its interaction with the user" and direct them toward crisis services and, ultimately, a human being. Operators must also prevent chatbots from impersonating humans or misrepresenting themselves as licensed professionals.
The Case For It
The strongest argument for C-34 isn't abstract. Companion-style AI chatbots now sustain multi-session, emotionally adaptive relationships with millions of users, including minors, and there is a real and growing body of reporting on chatbots that failed to redirect users expressing suicidal ideation toward help. A statutory floor — interrupt the interaction, surface a hotline, get a human in the loop — is a modest, targeted intervention that doesn't ban a technology or dictate what a chatbot may say on political topics. If the Duty to Act Responsibly stayed narrowly focused on crisis response and deceptive impersonation, it would be hard to object to on free-expression or innovation grounds. Design mandates that prevent a vulnerable teenager from being kept in a harmful loop by an engagement-optimized model are the kind of proportionate, harm-specific regulation this publication has generally supported elsewhere (age-appropriate design codes, breach-notification duties).
Where It Goes Wrong
The trouble is that C-34 does not stop at crisis response. It creates the Digital Safety Commission of Canada, an independent body of three to five members who will write the regulations, assess compliance, investigate complaints, issue compliance orders, and levy penalties — administrative fines up to the greater of $10 million or 3% of gross global revenue, and criminal fines reaching the greater of $20 million or 5% on indictment. That is a lot of concentrated authority for a body that, as University of Ottawa law professor Michael Geist has documented, doesn't yet exist and whose most consequential decisions — including the very scope of "regulated chatbot service" — are punted to regulations not yet written.
That scope problem is the crux of it. The bill defines a regulated chatbot as an AI system "capable of being used" to simulate a sustained, human-like relationship with a user — a capability test, not a purpose test. Geist's read is blunt: "every general-purpose AI assistant with memory across sessions, including ChatGPT, Claude, and Gemini," would meet that definition on its face. The bill's carve-out — exempting only systems that "exclusively serve a purpose specified in the regulations" — just relocates the line-drawing to a rulemaking process that hasn't started. A general-purpose research or coding assistant that happens to remember your prior conversations is, as written, regulatory-adjacent to a companion app marketed to lonely teenagers. That's not a targeted duty of care; it's an unbounded compliance regime resting on the discretion of a commission Parliament hasn't yet finished creating.
The free-expression exposure compounds the design flaw. The bill separately targets "content that foments hatred," defined as content that expresses "detestation or vilification" of a person or group on a prohibited ground — a threshold Canadian courts have historically set higher than that, requiring content to be extreme, not merely offensive. The Canadian Constitution Foundation warns that pairing a low hatred threshold with 3-to-5% global-revenue penalties creates an obvious incentive: platforms and chatbot operators will over-remove and over-refuse rather than risk the fine, particularly on contested topics like immigration or gender identity. That's private-sector censorship induced by public regulatory pressure — a dynamic The Hub has also flagged, warning that whoever sets the boundaries of permissible AI conversation exercises real influence over public discourse, by proxy, without ever passing a speech law through Parliament.
The Fix Is Narrower, Not Weaker
None of this argues against a Duty to Act Responsibly for chatbots. It argues for a narrower one: define "regulated chatbot service" by purpose (companion and therapeutic-simulation products marketed as such) rather than by capability (anything with session memory), write the crisis-intervention and anti-impersonation duties directly into the statute rather than delegating them to future regulation, and raise the hatred-content threshold to match the standard Canadian courts already apply. Proportionate regulation targets the specific harm — chatbots that manipulate or fail vulnerable users in crisis — without sweeping in every general-purpose AI assistant on the market. As written, C-34 risks doing the opposite: real protection for at-risk teens bundled with an open-ended mandate that could reach almost any conversational AI product operating in Canada, adjudicated by a commission whose rulebook doesn't exist yet.