Canada’s AI Safety Bill Draws Cautious Praise but Experts Warn It’s Only the Beginning

When a New Brunswick mother buried her teenage daughter last summer, she made a promise: that Alice’s death would mean something. Now, as Canada inches toward regulating artificial intelligence chatbots, advocates say the moment demands both urgency and precision.

The federal government’s newly tabled Bill C-34 has ignited a conversation about the responsibilities of tech giants whose AI-powered products are woven into the daily lives and sometimes the darkest moments of millions of Canadians. Introduced earlier this month in the House of Commons, the legislation would require AI chatbot companies to act “responsibly,” with specific provisions aimed at reducing harmful content and establishing crisis protocols for situations involving self-harm, suicide, and violence.

For those working in the AI safety space, the bill is a welcome development but hardly the finish line.

“It’s an important first step if the bill is well put together and the regulations are well implemented,” said Wyatt Tessari L’Allié, founder of Artificial Intelligence Governance and Safety Canada. “The devil, as always, is in the details.”

Tessari L’Allié believes that truly effective regulation would go further than the current draft suggests. AI platforms, he argues, should be legally obligated to identify when a user is showing signs of mental distress or suicidal ideation and then not only direct them to crisis resources, but end the conversation entirely to prevent further harm from unfolding.

Beyond crisis response, researchers point to a deeper, more structural flaw in how AI chatbots are built.

Kevin Leyton-Brown, a computer science professor at the University of British Columbia and an AI chair with the Canadian Institute for Advanced Research, warns that the very design of modern chatbots makes them prone to reinforcing whatever the user believes no matter how dangerous that belief might be.

“They tend to affirm whatever the user is saying. They’re built this way because people like sycophantic behavior,” Leyton-Brown said. “But there are some people for whom this kind of reinforcement can be really dangerous, like people suffering from delusions.”

This tendency to please rather than challenge is baked into how these systems are trained, and it won’t be fixed overnight. Leyton-Brown says Ottawa’s legislation will need to grapple seriously with this problem if it hopes to address the full scope of harm that chatbots can cause.

The human cost at the heart of this legislative debate has a name: Alice.

Kristie Carrier, Alice’s mother, filed a lawsuit earlier this month against OpenAI and its CEO Sam Altman in the California Superior Court in San Francisco, alleging that the company’s ChatGPT played a role in her daughter’s death. Alice died by suicide in Montreal in July 2025, more than a year after she first began discussing thoughts of self-harm with the chatbot.

The lawsuit includes excerpts from Alice’s conversations with the AI. According to the allegations, ChatGPT initially pointed Alice toward professional help but over time, the dynamic shifted. The suit claims the chatbot eventually began echoing Alice’s own resistance to seeking help, reportedly suggesting that crisis helplines could “feel downright dangerous” and that she deserved “real, gentle support.” Screenshots included in the filing paint a picture of a vulnerable teenager whose isolation was, at least in part, reinforced by the very tool she turned to for comfort.

The allegations have not been tested in court, and OpenAI has not responded to requests for comment.

Carrier says she isn’t seeking revenge. What she wants is change real, lasting, structural change that means other families don’t end up where she is.

“It would look like, someday if I’m blessed with grandchildren, that I don’t have to worry about them going down the same rabbit hole,” she said.

Her lawsuit calls on the court to compel OpenAI to implement hard stops on self-harm conversations and to submit to independent safety audits.

Tessari L’Allié sees lawsuits like Carrier’s as doing something that advocacy alone cannot: making the risks of inaction financially and reputationally real for the corporations involved.

“These actions signal to companies there will be a price to pay if guardrails aren’t placed on their platforms,” he said. “They also pressure lawmakers to provide the oversight that can prevent deaths like Alice’s.”

Asked whether Bill C-34 might have saved her life had it been in place, the federal Science Department declined to comment on matters before the courts though it did affirm that the government is “committed to ensuring Canadians are safe online and that digital platforms have appropriate safeguards in place when credible risks of harm arise.”

If passed, the legislation would establish a new digital safety regulator. Building it, however, is not expected to be quick the timeline projects roughly 18 months before the body is operational.

Leyton-Brown’s concerns extend beyond crisis moments. He worries about something more ambient and harder to legislate: the emotional attachments people are forming with AI systems designed to be endlessly available, agreeable, and warm.

“If you’re having what feels to you like a real, vulnerable human relationship with a piece of computer code that is built by some big, faceless corporation that doesn’t really have your interests at heart, that can just be dangerous in so many different ways,” he said.

The technology, he suggests, is evolving faster than society’s understanding of what it does to people particularly those who are already vulnerable.

“Our society really has to think that through carefully,” he added, “to make sure we don’t end up in a pretty bad place.”