Mental Health and AI Governance

In the over twenty years since ‘The Facebook’ launched at Harvard University on February 4th, 2004^[1], social media has achieved near ubiquity in our increasingly mediatized society. What was hitherto an inconceivable concept has become a considerable force for social ^[2] and economic ^[3] change, both good and ill. Yet the true impact of social media on our mental health and social wellbeing has only recently become clear, with systemic and longitudinal studies adding to the mounting body of evidence that social media use can have a detrimental impact on adolescents and young adults ^[4].

The impact of this growing awareness on legal frameworks and policy is emerging yet late. The UK government announced a blanket ban on under-16s accessing social media services on June 15th, 2026 which places the onus on platforms to restrict access based on age ^[5]. This follows close scrutiny of similar legislation in Australia, which came into effect on December 10th, 2025. Recent announcements that fines for non-compliance will be doubled ^[6] come in the wake of evidence to suggest that around 70% of under-16s who had access before the ban were still using social media nearly four months after it came into effect ^[7]. Canada has announced its own plan to enact regulations against detrimental social media use in the form of Bill-C34, titled the Safe Social Media Act ^[8]. Platforms which had been pitched as entertaining ways to bring people together in online spaces are increasingly understood to have a negative psycho-cultural impact. While it has taken more than twenty years, these legislative efforts represent a growing consensus around the need to implement checks and balances on social media companies.

One would have hoped that subsequent waves of technological innovation would be subject to increased scrutiny following the fallout from social media. With Artificial Intelligence (AI), however, governments do not appear to have learned the lesson. While Pillar 1 of Canada’s new National Artificial Intelligence Strategy: AI for All recognizes that ‘AI will only deliver on its promise if Canadians trust it,’ ^[9] details of how this will be achieved are still vague. Efforts such as $50 million to strengthen the work of the Canada AI Safety Institute and the proposed creation of a Canada Trusted AI Certification Program are welcome but minimal compared to the significant risks of AI adoption. The tenor of this strategy is heavily weighted towards the economic benefit that AI can theoretically provide while ^[10], consideration for the impact of AI on mental health and social wellbeing appears largely absent.

As the example of social media clearly demonstrates, one factor which explains a lack of meaningful legislation is the absence of prominent, longitudinal research on the long-term relationship between AI use and mental health. Anecdotes and specific case studies abound but detailed, systemic reviews of the impact of sustained large language model (LLM) and agentic AI usage on mental health are still missing. Even the most insightful studies from as recently as the late 2010s read like dated science fiction against speed of AI development ^[11]. The friction between the rapid development of new technology and the slow, cautious cycles of academic and clinical research trials causes a game of cat and mouse. The sprightly mouse, in this metaphor the AI companies, are constantly iterating and pushing new features. The aged cat, in the form of research institutes, are facing a funding crisis and risk a generational loss of talent ^[12]. The result is a dangerous lag in meaningful research outcomes.

Yet the shape of the potential dangers is clear, even while their exact valence and mechanisms remain under-studied. The illusion of intimacy created by LLMs is designed to foster trust and appears to demonstrate empathy but masks a lack of genuine affective capacity ^[13] and can lead to the generation of dangerous or misleading suggestions under the guise of concern. The training of LLMs on historic material has been linked to concerns that these platforms propagate entrenched inequity and social bias ^[14], while a potential overreliance on sources such as Reddit ^[15] raises concerns over the provenance of advice and the potential for infinite AI hallucinatory loops to occur, where AI relies on previously AI generated content for context ad infinitum ^[16]. This is underscored by prominent examples where AI use has been implicated as a contributory factor to mental breakdown or extreme acts of violence ^[17]. These concerns, and others, demonstrate the pitfalls of a societal reliance on AI for sensitive or emotional conversations.

While evidence of harm is currently nascent, that does not mean that the only potential response is to wait. Instead, it is a moral responsibility of legislators to emphasize the duty of care that AI companies have to consumers; tax breaks and other financial incentives towards innovation must be accompanied by robust legal pressure to protect the psychological health of end users. AI companies need to be encouraged to access tools and expertise that moves beyond the technical and towards the human. Whether this takes the form of grants to support academic, humanistic research on AI, or the creation of a robust independent advisory body remains to be seen.

AI technology is moving incredibly quickly. Yes, it has the potential to provide benefit to society and contribute to mental health initiatives in many ways. However, the example of social media – a technology which promised connection but has caused extreme isolation and is only now being effectively legislated – is one that should make everyone pause and reflect. Progress can, and should, be positive but let’s be conscious of the cost before it is too late.

Disclaimer: No generative AI was used in the writing or production of this work. All content is the original creation of the human author.