OTTAWA — The federal government must toe a delicate line between safety and privacy if it chooses to fold AI platforms into online safety laws, say advocates who warn that Canadians’ intimate exchanges with AI chatbots could be exposed in efforts to prevent future tragedies. 

Federal ministers began publicly discussing the prospect of regulating AI-driven chatbots after they learned 18-year-old Jesse Van Rootselaar shared concerning messages with ChatGPT in June that OpenAI didn’t report to police until February, after Van Rootselaar shot and killed eight people before taking her own life in Tumbler Ridge, B.C.

AI Minister Evan Solomon has demanded details about the safety protocols of AI platforms operating in Canada, and Justice Minister Sean Fraser threatened to legislate changes if they don’t meet the government’s expectations. Several experts have urged the government to include chatbots in the online harms bill it plans to table this year to regulate social media platforms.

After meeting with ministers, OpenAI said that if new protocols had been in place before the shootings, Van Rootselaar’s messages would have been reported to police. In response to the government’s concerns, the company promised lessons learned from Tumbler Ridge would inform its policies on when to escalate concerns to police, and said it would establish points of contact with Canadian law enforcement agencies. 

Solomon said he still doesn’t have enough details about the proposals, and is expected to meet with OpenAI’s CEO Sam Altman to get more clarity.

Amid all the stern talk over the last week, though, there’s been little debate about the potential dangers of regulating AI platforms in the same way as social media.

“It’s incredibly delicate for both the government and the companies, because there’s only so much that the government could set down in law that would be appropriate,” said Emily Laidlaw, Canada Research Chair in cybersecurity law and associate law professor at the University of Calgary. 

Laidlaw is a member of a Trudeau-era expert advisory group on online safety, which the Heritage Department reconvened last week to weigh in on extending proposed online harms laws to AI chatbots. 

Unlike social media content, which is often public-facing, conversations with chatbots tend to be seen by users as private—like writing in a diary or confiding in a friend. People consult AI for medical questions, relationship advice and financial information, among other matters.  

Set the safety thresholds too high, and those intimate conversations could be shared at the slightest suspicion of illegal activity, said Laidlaw. Too low, and they may fail to stop future tragedies like the one in Tumbler Ridge. 

Laidlaw supports including AI platforms in the legislation, but is candid about the scale of the challenge. “The government, if they are going to play a role here, are going to have to very narrowly set down what that threshold would be,” she said. 

For now, it’s up to companies to figure it out themselves. It’s not clear exactly what Van Rootselaar said to ChatGPT in June that caused the platform to flag her messages to the company’s safety team, but The Wall Street Journal reported that she had described scenarios involving gun violence. 

OpenAI staff discussed the chat, but decided not to report it to police because they didn’t feel there was a credible and immediate threat of real-world violence. “OpenAI works continuously to identify potential warning signals of serious violence, while also protecting the privacy and security of the vast majority of people who use our tools responsibly,” Ann O’Leary, the company’s vice-president of global policy, said in a letter to Solomon last week. 

Canadians appear to have little trust in companies to make judgments when it comes to their privacy, though. A survey commissioned by the federal privacy commissioner’s office one year ago suggested that 89 per cent of Canadians were at least somewhat concerned about their personal privacy, and only 40 per cent said they believe businesses respect their privacy rights. Faith in the government to respect their privacy rights was somewhat higher, at 62 per cent.

When it comes to AI in particular, the vast majority of Canadians, 83 per cent, are worried about their privacy, according to a Leger poll conducted last August. 

Legislating changes may not help, said Ann Cavoukian, the former privacy commissioner of Ontario, who now advises companies and governments on using technology to protect privacy. Once able to demand personal information be shared, governments will always find a reason why doing so would benefit the public good, she said. 

“Anytime there are these horrendous events that take place, like Tumbler Ridge, all of a sudden safety is ratcheted up—the importance of safety and protecting people,” said Cavoukian. “It reduces the importance of privacy and personal information because there’s always some other reason why it’s more important to gain access to your data. That’s wrong.”

Finding a compromise will be essential to getting the legislation through Parliament. Conservatives called for an independent inquiry into the Tumbler Ridge shooting that would look into the AI used by the shooter, and the relationship between ethical use of AI and individual privacy. The party wants online service providers, including social media platforms and AI chatbots, held to a duty of care of their users, said Conservative MP Michelle Rempel Garner, but not at the expense of their privacy.

“My key concern with the previous versions of the government’s online harms bill has been that they ask Canadians to sacrifice their civil liberties in order to keep them safe online,” Rempel Garner said in an interview. “I’m so concerned that they haven’t tried to find a way forward that isn’t just looking at restrictions on speech or incursions on privacy.” 

Rempel Garner tabled her own vision for an online harms bill in Parliament last year. It would cover AI platforms, but require a judge rather than a government body to decide whether personal information should be accessed. Culture Minister Marc Miller told The Logic last month he would be interested in consulting with Rempel Garner to see if parts of her bill could be adopted or included in the government’s plan.

Privacy concerns are legitimate, said Taylor Owen, the founding director of McGill University’s Centre for Media, Technology and Democracy, and a member of the government’s expert advisory group. But, as the Tumbler Ridge shooter’s conversations demonstrate, companies are already monitoring peoples’ conversations, he noted. 

“It’s already happening, in the absence of regulation, anyway. We just don’t know how and by whom and under what terms it’s being used,” Owen said.