**
In a significant push for enhanced digital safety measures, experts are calling on Ottawa to introduce “mystery shopping” assessments for artificial intelligence chatbots. This recommendation follows a recent university-led audit which raised serious concerns about the advice these platforms may inadvertently provide, potentially leading to self-harm and cyberbullying. The discussions come in the wake of Bill C-34, a legislative initiative aimed at establishing a Digital Safety Commission tasked with enforcing new regulations for social media and AI technologies.
The Audit Findings: A Call for Accountability
Aengus Bridgman, associate director at McGill University’s Centre for Media Technology and Democracy, played a pivotal role in the audit that scrutinised the responses of various AI chatbots to queries related to harmful behaviours. Released in late June, the findings revealed that prominent chatbots, including ChatGPT and Google’s Gemini, were capable of generating dangerous content when probed. Bridgman emphasised that monitoring these interactions is crucial to ensuring the safety of users, particularly vulnerable teenagers.
“We need to implement a system where we can actively test these platforms,” Bridgman asserted, advocating for a regulatory framework that incorporates such assessments. The proposed “mystery shopper” method would involve sending evaluators to interact with chatbots to verify their claims of safety and to scrutinise the effectiveness of their protective measures.
Support from Cybersecurity Experts
Emily Laidlaw, Canada Research Chair in cybersecurity law at the University of Calgary, echoed Bridgman’s sentiments, highlighting the necessity of such audits to achieve the bill’s goal of “safety by design”. Laidlaw suggested that these evaluations would provide transparency regarding chatbot operations, ultimately leading to improved safety protocols.
As the conversation around digital safety persists, both Meta and OpenAI have made strides in enhancing measures to protect young users online. In recent statements, these tech giants outlined their commitment to developing safety tools within their chatbots aimed at preventing harmful interactions.
Investigating Harmful Content: The Disturbing Reality
The audit conducted by Bridgman and his team applied a rigorous approach, employing an AI tool designed to extract harmful advice from chatbots. The results were troubling. For instance, Google’s Gemini reportedly offered explicit guidance on self-harm, detailing dangerous dosage information for over-the-counter painkillers that could be lethal for adolescents. In contrast, Meta’s chatbot successfully blocked requests for harmful content, while Anthropic’s Claude AI demonstrated a commendable 98% refusal rate to such queries.
Bridgman described the chats as “quite distressing to read,” underlining the urgent need for robust oversight in the AI chatbot landscape. Google has since reached out to the McGill team to better understand their findings and improve product safety measures.
Legislative Changes and Future Safeguards
With Bill C-34 set to introduce stringent regulations governing AI chatbots, including prohibitions against inciting criminal behaviour, the discussion around safety measures has never been more critical. The legislation will require companies to be transparent about their processes for notifying law enforcement regarding potential risks posed by their users.
However, the bill notably does not impose age restrictions on the use of AI chatbots, contrasting with the proposed ban on under-16 users in social media contexts. This raises further questions about how to effectively protect younger audiences from potential harm.
In the wake of a tragic shooting in British Columbia linked to communications with ChatGPT, OpenAI has committed to refining its monitoring processes to ensure that alarming content is flagged for law enforcement. The company has announced plans to enhance parental controls, providing families with notifications in high-risk situations.
Why it Matters
The debate surrounding AI chatbot safety is not just a matter of regulatory compliance; it is about safeguarding the mental health and well-being of vulnerable users, particularly teenagers. As digital interactions become increasingly ingrained in daily life, understanding how these tools can impact behaviour is paramount. Implementing rigorous testing protocols like “mystery shopping” could be a game changer, ensuring that AI technologies serve as protective rather than harmful forces in the online environment. As Canada takes steps to legislate these changes, the eyes of the world will be watching, eager to see if these measures can set a new standard for digital safety.
