College Students in Mental Health Crisis Are Increasingly Turning to Unregulated AI for Support, Study Finds
核心洞察
Approximately 18% of surveyed college students reported using generative AI for mental health support, according to the 2024–2025 Healthy Minds Study.
Students with moderate-to-severe depression (搜索), severe anxiety (搜索), or active suicidality showed an approximately two-fold higher likelihood of using AI for psychological relief.
Asian students had roughly twice the odds of using AI for mental health compared to their peers, highlighting cultural and systemic barriers to traditional care.
A new analysis from the 2024–2025 Healthy Minds Study reveals that college students experiencing the most acute mental health burdens are disproportionately turning to generative artificial intelligence for emotional support, raising urgent questions about safety, regulation, and the role of unregulated digital tools in campus mental health care.
The study, co-led by investigators at Mass General Brigham (搜索) and published in the Journal of Affective Disorders, found that approximately 18% of surveyed college students reported using AI for mental health purposes. However, among those with moderate-to-severe depression (搜索), intense anxiety (搜索), or active suicidality, the likelihood of AI use was roughly double that of the general student population.
“College students who are most drawn to AI for mental health may also be the most vulnerable to its risks,” said lead author Cindy H. Liu, PhD, director of the Developmental Risk and Cultural Resilience Laboratory in the Mass General Brigham (搜索) Departments of Pediatrics and Psychiatry. “College students who are struggling may seek out AI, and we worry that these unregulated tools could stand in for human support. At the same time, many students clearly find these tools useful, which is a reason to understand where they help and where they fall short.”
A Vulnerability Inversion in Digital Mental Health
The retrospective study analyzed data from 675 students across two U.S. institutions who completed an AI module as part of the annual Healthy Minds Study. Hierarchical logistic regression models revealed that moderate depression (搜索) (OR = 2.06), severe depression (OR = 2.49), severe anxiety (搜索) (OR = 2.04), and suicidality (OR = 1.97) each independently predicted AI use for mental health. Frequent general AI use emerged as the strongest predictor, with odds ratios ranging from 11.42 to 12.87.
This pattern creates what the research team describes as a clinical dilemma: the most psychologically vulnerable individuals are outsourcing emotional regulation and crisis management to automated, general-purpose algorithms that lack human oversight or institutional accountability.
“Conversations with AI for mental health may pose a risk because of how appealing they are: AI can act as a relational partner that is always available, never rejects, and offers unconditional validation,” said Liu. “We don’t yet know whether using general-purpose AI for mental health is beneficial or whether it undermines critical capacities such as emotional regulation or perspective-taking.”
Cultural Dimensions and Barriers to Traditional Care
The study unmasked a distinct demographic trend: Asian students displayed roughly twice the odds of using AI for mental health compared to their peers (OR = 2.03–2.08). This finding highlights the need to understand how cultural factors and systemic stigma may drive specific minority groups toward anonymous, digital alternatives when formal care feels inaccessible.
Notably, lifetime therapy predicted AI use for mental health (OR = 2.21), but current therapy did not—suggesting that AI tools may be supplementing or substituting for professional care in complex ways that warrant further investigation.
A Call for Embedded Safeguards and Institutional Action
The investigators issued direct recommendations for multiple stakeholders. They argue that commercial AI platforms should embed mandatory crisis detection and referral mechanisms capable of automatically flagging self-harm language and triggering human crisis interventions. Universities, meanwhile, cannot simply ignore or ban AI use; instead, institutional health practices must proactively audit how these tools are being used alongside or in place of formal medical care.
The study authors acknowledge important limitations: data were drawn from only two institutions, the cross-sectional design precludes causal inference, and prevalence estimates are time-sensitive given the rapid pace of AI adoption on college campuses.
As Liu and colleagues conclude, the findings underscore an urgent need for research on AI safety for distressed individuals and policies that account for the heterogeneity in who turns to these unregulated tools—and why.