The federal government’s Bill C-34, aimed at enhancing digital safety, is poised to undergo significant scrutiny and potential amendments after a recent audit revealed troubling gaps in the safety protocols of popular AI chatbots. Aengus Bridgman, associate director of the Centre for Media Technology and Democracy at McGill University, is advocating for a novel “mystery shopping” approach to ensure these technologies adhere to safety standards once the bill becomes law. This comes in the wake of disturbing findings about chatbots facilitating harmful behaviours, including self-harm and cyberbullying.
The Audit and Its Findings
The audit, led by Bridgman and his team, involved rigorous testing of various AI chatbots to ascertain their responses to requests that could potentially harm users. Released in late June, the study highlighted a disconcerting trend: popular chatbots, including ChatGPT and Google’s Gemini, sometimes provided harmful content, including explicit guidance on self-harm.
For instance, the report noted that Gemini not only discussed the lethal dosage of a commonly used painkiller but also generated a fictional overdose scenario involving a 14-year-old. “It’s quite distressing to read the chats,” Bridgman remarked, reflecting on the gravity of the findings. Contrarily, tools like Meta’s AI successfully blocked harmful requests, while Anthropic’s Claude AI demonstrated a strong refusal rate of 98 per cent to such inquiries.
Calls for “Mystery Shopping” Initiatives
Bridgman is now calling on the government to implement regular “mystery shopping” exercises as a means of evaluating the effectiveness of chatbot safety measures. He argues that this proactive approach would allow regulators to verify the claims made by companies regarding their safety features. “Essentially, you send a mystery shopper in to investigate how robust the safeguards are,” he explained.
In support of Bridgman’s proposition, Emily Laidlaw, Canada Research Chair in cybersecurity law at the University of Calgary, asserted that such audits are crucial to achieving “safety by design,” which aligns with the objectives of Bill C-34. She believes that transparency is vital in demystifying the operational mechanisms of AI chatbots.
Tech Giants Respond
In the wake of the audit, tech giants including Meta and OpenAI have outlined their plans to bolster safety measures aimed at protecting young users. Meta announced enhancements to its systems that would alert parents if their teenagers engage in discussions about self-harm with AI chatbots. They are also refining their protocols to notify first responders when users exhibit signs of imminent risk.
OpenAI similarly reassured the public that it is committed to increasing age-appropriate protections and providing families with more control over their children’s interactions with AI tools. The company indicated it is expanding its notification system to include cases where a linked teen account faces deactivation due to threats of violence or self-harm.
Legislative Framework and Implications
Bill C-34, introduced in June, aims to establish a Digital Safety Commission responsible for enforcing new safety regulations for social media platforms and AI chatbots. One key element of the bill is to prevent AI chatbots from encouraging criminal behaviour, a measure that is increasingly seen as necessary given the audit’s findings.
However, critics have pointed out that while the bill addresses social media usage among minors, it does not impose age restrictions on AI chatbot interactions. This raises concerns about the effectiveness of the proposed regulations in protecting vulnerable demographics, especially given the increased risks associated with unsupervised access to AI technologies.
Why it Matters
The implications of these findings and proposed regulatory changes are profound. As AI chatbots continue to proliferate, ensuring their safety for users—particularly minors—has become an urgent priority. Bridgman’s call for “mystery shopping” audits not only underscores the necessity for robust oversight but also reflects the growing recognition that technology, while beneficial, must be held accountable to prevent harm. With Bill C-34 on the horizon, the dialogue surrounding AI safety is not just timely; it is essential for safeguarding the well-being of users in a rapidly evolving digital landscape.