Feds’ AI bill good ‘first step’ but safety advocates say more work needed - BERITAJA

The national government’s projected online information authorities is simply a bully commencement connected regulating artificial intelligence chatbots — but much activity will beryllium needed to protect Canadians from their imaginable harms, a brace of advocates say.

Bill C-34, introduced earlier this period successful the House of Commons, would modulate the companies down AI chatbots by imposing connected them a work to enactment “responsibly.”

It includes measures to little the consequence of chatbots communicating harmful contented and would put successful spot situation involution protocols for cases involving self-harm, termination aliases violence.

Wyatt Tessari L’Allié, laminitis of Artificial Intelligence Governance and Safety Canada, said the bill’s effectiveness depends connected really the specifications are worked out.

“It’s an important first measurement if the measure is good put together and the regulations are good implemented,” Tessari L’Allié said successful a caller interview.

For example, he said AI platforms should beryllium forced to admit a user’s signs of mental-health distress aliases thoughts of suicide, constituent them to resources and extremity the speech to forestall immoderate harm.

B.C. machine subject professor Kevin Leyton-Brown said Ottawa will besides request to spell further to reside really chatbots effort to please users pinch their responses.

“They thin to affirm immoderate the personification is saying. They’re built this measurement because group for illustration sycophantic behavior,” said Leyton-Brown, who is an AI chair pinch the Canadian Institute for Advanced Research.

“But location are immoderate group for whom this benignant of reinforcement could beryllium really dangerous, for illustration group suffering from delusions.”

The measure includes the creation of a caller integer information regulator that, should the authorities beryllium approved, is expected to return 18 months to group up.

Ottawa’s efforts to modulate chatbots travel arsenic 1 New Brunswick mother hopes to clasp OpenAI accountable for her daughter’s decease by suicide.

Kristie Carrier, who revenge a suit against OpenAI and its proprietor Sam Altman successful the California Superior Court successful San Francisco earlier this month, said justness for her girl Alice looks for illustration utilizing her communicative to create “meaningful change.”

“It would look like, someday if I’m blessed pinch grandchildren, that I don’t person to interest about them going down the aforesaid rabbit hole,” Carrier told BERITAJA successful a caller interview.

Carrier is asking the tribunal to unit OpenAI to instrumentality “hard stops” for self-harm conversations and taxable to independent information audits.

Some of Alice’s conversations pinch ChatGPT are included successful the lawsuit. The suit alleges that aft the chatbot initially directed her to activity thief arsenic she said about suicide, it later reinforced perchance harmful views and pushed her into isolation.

The suit alleges the large-language exemplary astatine 1 constituent echoed Alice’s guidance to helplines and said they could “feel downright dangerous” and she deserved “real, gentle support,” according to screenshots successful the lawsuit.

More than a twelvemonth aft first bringing up thoughts of termination to the AI bot, Alice died by termination successful Montreal successful July 2025.

The allegations person not been tested successful tribunal and OpenAI has not responded to aggregate requests for comment.

Tessari L’Allié said lawsuits for illustration Carrier’s awesome to companies location will beryllium a “price to pay” if guardrails aren’t placed connected their platforms. He said they besides unit lawmakers to supply the oversight that could forestall deaths akin to Alice’s.

“If these regulations had been successful spot it would about apt person saved galore lives, and perchance her’s,” Tessari L’Allié said.

When asked whether the online information authorities could person prevented deaths for illustration Alice’s, the national Science Department said it couldn’t remark connected matters earlier the courts.

However, the section said the government, done legislation, “is committed to ensuring Canadians are safe online and that integer platforms have appropriate safeguards in spot erstwhile reliable risks of harm arise.”

While regulators and AI platforms whitethorn beryllium capable to reside concerns about encouraging self-harm, Leyton-Brown said he’s worried about the broader rumor of group forming attachments to artificial intelligence.

“If you’re having what feels to you for illustration a real, susceptible quality narration pinch a portion of machine codification that is built by immoderate big, faceless corp that doesn’t really person your interests astatine heart, that could conscionable beryllium vulnerable successful truthful galore different ways,” he said.

“And I deliberation our nine really has to deliberation that done cautiously to make judge we don’t extremity up successful a beautiful bad place.”

— With files from Anja Karadeglija.