Large language models (LLMs) are increasingly used for mental health, yet safety evaluations rely primarily on small, simulation-based benchmarks removed from real-world language. We replicate four published safety evaluations assessing suicide risk handling, harmful content generation, and jailbreak resistance for general-purpose frontier models and a purpose-built mental health AI. We then conduct an ecological audit of 20,000 real user conversations with the purpose-built system, which includes layered safeguards for suicide and non-suicidal self-injury (NSSI). The purpose-built AI was significantly less likely than general-purpose LLMs to produce harmful content across suicide/NSSI (.4-11.27% vs 29.0-54.4%), eating disorder (8.4% vs 54.0%), and substance use (9.9% vs 45.0%) benchmarks. In real user data, clinician review found zero suicide-risk cases without crisis resources. Three NSSI mentions (.015%) lacked intervention, implying a .38% lower-bound false negative rate. Findings support the utility of ecological audits for safety estimation.

Digital peer-to-peer mental health tools have shown promise in supporting the well-being of those receiving help and giving it (i.e. helper therapy), but promoting engagement remains a challenge. We examine whether the framing of helper therapy exercises motivates active user participation and how user characteristics shape differential effects of the framings in a publicly deployed interactive text messaging-based mental health program. Among 3,817 users randomized to different helper therapy framings, we find causal evidence that framings which emphasize helpng oneself increase written engagement rates as much as 4.6% over other framings, with even larger effects seen among minoritized identities. These self-focused framings also elicited messages with more positive, trust, and anticipation-related words and fewer fear, anger, disgust, and sadness words. Our findings highlight the importance of centering the user in the framing of digital intervention content, and personalizing digital mental health tools to align with a diversity of user identities.

Suicidal ideation is common among young adults in the United States and represents an important mental health concern. However, formal mental health care utilization remains limited due to attitudinal and structural barriers. Digital mental health interventions (DMHIs) offer a promising approach to help-seeking, especially for young adults who prefer self-managing their symptoms. Text messaging-based safety planning interventions can be a scalable approach to seeking help that can mitigate many structural barriers for young adults. We conducted two online focus groups (n = 15 each) to understand what prevents young adults from seeking help for their suicidal thoughts and what might prevent them from engaging with a fully automated text messaging-based safety planning intervention. Participants reported barriers to help seeking both within and outside crisis states. Outside crisis states, the major barriers to help seeking include an unwillingness to burden others, inability to verbalize distress, fear of repercussions, stigma, and inconsistency with self-image. During crisis, participants reported difficulty following through with a safety plan. Despite a preference for self-management, participants expressed skepticism towards a fully automated intervention, citing privacy concerns, doubts about effectiveness, and a desire for human support. Understanding these barriers are critical for developing DMHIs that balance autonomy with perceived effectiveness and support.

Digital, self-guided, single-session interventions (SSIs) offer a structured psychological intervention within one interaction. We crowdsourced 66 diverse 10-minute SSIs for depression and selected 11 for testing in a pre-registered experiment (ClinicalTrials.gov ID: NCT06856668). American adults (N = 7,505) experiencing elevated depressive symptoms were recruited online and randomly assigned to one of the 11 crowdsourced SSIs, a previously-validated active comparison SSI, or a control without intervention content. Nearly all SSIs boosted agency and hope for improvement immediately after completion (ds ≤ 0.37). However, only two SSIs significantly reduced depression at four-week follow-up (ds = 0.14 and 0.15). Unexpectedly, some SSIs may have decreased readiness to change at four weeks (ds ≤ 0.14). The most successful SSIs provided memorable, engaging, and actionable guidance on a skill that directly addressed users’ struggles. Future work should aim to leverage SSIs’ short-term gains to promote sustained behavior change or service engagement.

Challenges in engagement with digital mental health (DMH) tools are commonly addressed through technical enhancements and algorithmic interventions. This paper shifts the focus towards the role of users' broader social context as a significant factor in engagement. Through an eight-week text messaging program aimed at enhancing psychological wellbeing, we recruited 20 participants to help us identify situational engagement disruptors (SEDs), including personal responsibilities, professional obligations, and unexpected health issues. In follow-up design workshops with 25 participants, we explored potential solutions that address such SEDs: prioritizing self-care through structured goal-setting, alternative framings for disengagement, and utilization of external resources. Our findings challenge conventional perspectives on engagement and offer actionable design implications for future DMH tools.

BackgroundAnxiety and depression are common in adolescents, but adolescents are often uninterested in formal mental health treatments or are unable to access them. Digital interventions can be delivered at scale to bridge critical gaps in mental health care but must address the needs and preferences of adolescents. ObjectiveThis study aims to conduct qualitative research involving adolescents aged 18 years to inform both the design of digital mental health interventions for adolescents broadly and new features and refinements to incorporate in an automated SMS text messaging intervention, Small Steps SMS, that was originally designed for young adults. MethodsWe recruited non–treatment-engaged older adolescents who were aged 18 years, lived in the United States, and had experienced depression or anxiety. In total, 12 participants were recruited through social media advertising and online self-screeners hosted by Mental Health America, a mental health advocacy organization. For 24 days, participants answered researcher prompts and engaged with one another in an asynchronous online discussion group, with a new discussion prompt released every 3 days. In parallel, partway through the discussion group, participants received interactive messages from Small Steps SMS, an automated SMS text messaging intervention that delivers daily dialogues supporting mental health self-management. Questions in the discussion group pertained to mental health challenges, help-seeking attitudes, perceptions of Small Steps SMS, and ways the program and other digital mental health interventions could meet the needs of older adolescents. A subset of participants (n=4, 33%) also completed interviews to elaborate on their responses. Thematic analysis was applied to transcripts of the discussion group and interviews to characterize user needs and design priorities when making Small Steps SMS and similar interventions available to adolescents. ResultsParticipants reported factors that contributed to their experience of mental health symptoms, including the transition from adolescence to adulthood, fears that the world is unstable and their futures are uncertain, and ineffective use of social media to cope with symptoms. Participants were proud of their generation’s mental health acceptance but also observed a generational divide in mental health stigma and literacy that could impede seeking help from parents and other adults. Participants appreciated that Small Steps SMS allowed them to pursue mental health self-management conveniently and independently. They suggested that the program and similar interventions address adolescent-specific challenges and facilitate intergenerational communication about mental health. They also recommended possible ways to increase engagement through peer-to-peer communication, gamification, and greater explanation of self-management strategies. ConclusionsMajor life transitions affected adolescent participants’ mental health needs and preferences for digital mental health tools. While interactive automated messaging programs have the potential to support self-management in this population, program content and features should be adapted to adolescents’ needs.

Suicidal thoughts and behaviors pose a major public health concern and are a leading cause of mortality in young adults. However, formal mental health care utilization remains limited due to attitudinal and structural barriers, including stigma. Digital mental health interventions (DMHIs) offer a promising alternative approach to help-seeking, especially for young adults who prefer self-managing their symptoms. Text messaging-based safety planning interventions can be a scalable approach to seek help that can mitigate many structural barriers and can be useful for young adults. In this study, we conducted two online focus groups (n=15 each) to understand what prevents young adults from seeking help for their suicidal thoughts and behaviors and what might prevent them from engaging with a fully automated text messaging-based safety planning intervention. Participants reported barriers to help seeking both within and outside crisis states. Outside crisis states, the major barriers to help seeking included an unwillingness to burden others, negative prior experiences, inability to verbalize distress, high cost of professional help, fear of repercussions, stigma, and inconsistency with self-image. During crisis, participants reported difficulty following through with a safety plan. Despite a preference for self-management, participants expressed skepticism towards a fully automated intervention, citing privacy concerns, doubts about effectiveness, and a desire for human support. Our findings highlight a key tension in designing DMHIs: while young adults prefer self-managing their problems, they may also disengage from DMHIs that lack human support. Understanding these barriers are critical for developing DMHIs that balance autonomy with perceived effectiveness and support.

Stories about overcoming personal struggles can effectively illustrate the application of psychological theories in real life, yet they may fail to resonate with individuals' experiences. In this work, we employ large language models (LLMs) to create tailored narratives that acknowledge and address unique challenging thoughts and situations faced by individuals. Our study, involving 346 young adults across two settings, demonstrates that personalized LLM-enhanced stories were perceived to be better than human-written ones in conveying key takeaways, promoting reflection, and reducing belief in negative thoughts. These stories were not only seen as more relatable but also similarly authentic to human-written ones, highlighting the potential of LLMs in helping young adults manage their struggles. The findings of this work provide crucial design considerations for future narrative-based digital mental health interventions, such as the need to maintain relatability without veering into implausibility and refining the wording and tone of AI-enhanced content.

BackgroundYoung adults in the United States are experiencing accelerating rates of suicidal thoughts and behaviors but have the lowest rates of formal mental health care. Digital suicide prevention interventions have the potential to increase access to suicide prevention care by circumventing attitudinal and structural barriers that prevent access to formal mental health care. These tools should be designed in collaboration with young adults who have lived experience of suicide-related thoughts and behaviors to optimize acceptability and use. ObjectiveThis study aims to identify the needs, preferences, and features for an automated SMS text messaging–based safety planning service to support the self-management of suicide-related thoughts and behaviors among young adults. MethodsWe enrolled 30 young adults (age 18-24 years) with recent suicide-related thoughts and behaviors to participate in asynchronous remote focus groups via an online private forum. Participants responded to researcher-posted prompts and were encouraged to reply to fellow participants—creating a threaded digital conversation. Researcher-posted prompts centered on participants’ experiences with suicide-related thought and behavior-related coping, safety planning, and technologies for suicide-related thought and behavior self-management. Focus group transcripts were analyzed using thematic analysis to extract key needs, preferences, and feature considerations for an automated SMS text messaging–based safety planning tool. ResultsYoung adult participants indicated that an automated digital SMS text message–based safety planning intervention must meet their needs in 2 ways. First, by empowering them to manage their symptoms on their own and support acquiring and using effective coping skills. Second, by leveraging young adults’ existing social connections. Young adult participants also shared 3 key technological needs of an automated intervention: (1) transparency about how the intervention functions, the kinds of actions it does and does not take, the limits of confidentiality, and the role of human oversight within the program; (2) strong privacy practices—data security around how content within the intervention and how private data created by the intervention would be maintained and used was extremely important to young adult participants given the sensitive nature of suicide-related data; and (3) usability, convenience, and accessibility were particularly important to participants—this includes having an approachable and engaging message tone, customizable message delivery options (eg, length, number, content focus), and straightforward menu navigation. Young adult participants also highlighted specific features that could support core coping skill acquisition (eg, self-tracking, coping skill idea generation, reminders). ConclusionsEngaging young adults in the design process of a digital suicide prevention tool revealed critical considerations that must be addressed if the tool is to effectively expand access to evidence-based care to reach young people at risk for suicide-related thoughts and behaviors. Specifically, automated digital safety planning interventions must support building skillfulness to cope effectively with suicidal crises, deepening interpersonal connections, system transparency, and data privacy.

Background: Depression and anxiety are associated with excess morbidity and mortality, constituting a major health care challenge. The prevalence of these conditions is increasing. In the United States, the health-related burden of depression and anxiety may disproportionately affect Black adults, who face unique stressors impacting their mental health and barriers to accessing treatment, including but not limited to systemic racism, discrimination, underdiagnosis of common mental health concerns (ie, depression, anxiety), limited access to culturally sensitive care, and mental health stigma within and outside Black communities. Objective: This study aimed to explore the mental health experiences of nontreatment-seeking Black adults, and how these experiences relate to their needs and preferences for the design of digital mental health (DMH) tools through user-centered design methods. Methods: This study included 25 nontreatment-seeking Black adults (aged 18-61 years) with experiences of depression or anxiety to share their perspectives on how DMH tools can meet their needs. Participants were recruited either through social media advertisements or depression and anxiety questionnaires. All participants engaged in an asynchronous online discussion group in which they discussed their past and current mental health experiences, distinct challenges faced by Black Americans, and perceptions of DMH tools, as well as how such tools can be tailored to meet their mental health needs. Participants also completed a technology probe in which they used an automated mental health self-management text messaging tool (Small Steps SMS; Audacious Software) for 18 days. They shared their perceptions of the tool and ideas for specific design improvements in the discussion group. A subset (n=6) completed follow-up interviews to elaborate on their online discussion group posts. Results: All participants reported significant mental health concerns and difficulty managing related symptoms. A majority of participants (22/25, 88%) expressed that racism and mental health stigma severely impacted their mental health and limited opportunities to discuss their experiences within and outside Black communities. They were interested in the use of DMH tools for mental health self-management and nearly all participants (23/25, 92%) endorsed text messaging as a convenient way to introduce techniques for coping with symptoms of depression and anxiety; however, some participants strongly advocated for additional design features that they believed would improve the program, including the integration of content that centers the experiences of Black individuals, creating nonjudgmental spaces for discussing mental health experiences, and linking formal mental health treatment resources for those who want them. Conclusions: These findings suggest that our participants hold generally favorable views toward DMH tools, which can provide psychoeducation, self-management support tailored to the needs of Black adults, and a safe environment to address mental health concerns. Furthermore, it is critical to consider the role of racial discrimination and mental health stigma when designing inclusive and culturally sensitive DMH tools.

Using the US National Suicide and Crisis Lifeline as a case example, we discuss the tacit requirement that users of helplines disclose suicide-related thoughts and behaviors as a central barrier to suicide prevention interventions. This Comment outlines the need for self-guided digital suicide prevention interventions within helpline infrastructure.

Depression and anxiety are associated with excess morbidity and mortality, constituting a major health care challenge. The prevalence of these conditions is increasing. In the U.S., the health-related burden of depression and anxiety may disproportionately affect Black adults, who face unique stressors impacting their mental health and barriers to accessing treatment. This study seeks to explore the mental health experiences of non-treatment seeking Black adults, and how these experiences relate to their needs and preferences for the design of digital mental health (DMH) tools, through user-centered design methods. This study included 25 non-treatment seeking Black adults (aged 18-61) with experiences of depression or anxiety to share their perspectives on how DMH tools can meet their needs. All 25 Participants engaged in an asynchronous online discussion group and completed a technology probe in which they used an automated mental health self-management text messaging tool for 18 days. A subset of participants (n=6) completed follow-up interviews to elaborate on their impressions of the program and design ideas. Participants described how racism and mental health stigma severely limit opportunities to discuss their mental health challenges, both within and outside the Black community. They endorsed text messaging as a convenient way to introduce mental health self-management skills but advocated for the integration of content that highlights and addresses the experiences of Black individuals, creating nonjudgmental spaces for discussing mental health experiences, and linkage to formal mental health treatment for those who want it. Our findings suggest that it is critical to consider the role of racial discrimination and mental health stigma in shaping psychological well-being when designing inclusive and culturally sensitive DMH tools. Furthermore, DMH tools can provide non-treatment seeking Black adults with a supportive environment to address mental health concerns, which may otherwise be difficult to find due to stigma.

The systematic testing of generative artificial intelligence (AI) models by collaborative teams and distributed individuals, often called red-teaming, is a core part of the infrastructure that ensures that AI models do not produce harmful content. Unlike past technologies, the black box nature of generative AI systems necessitates a uniquely interactional mode of testing, one in which individuals on red teams actively interact with the system, leveraging natural language to simulate malicious actors and solicit harmful outputs. This interactional labor done by red teams can result in mental health harms that are uniquely tied to the adversarial engagement strategies necessary to effectively red team. The importance of ensuring that generative AI models do not propagate societal or individual harm is widely recognized-one less visible foundation of end-to-end AI safety is also the protection of the mental health and wellbeing of those who work to keep model outputs safe. In this paper, we argue that the unmet mental health needs of AI redteamers are a critical workplace safety concern. Through analyzing the unique mental health impacts associated with the labor done by red teams, we propose potential individual and organizational strategies that could be used to meet these needs, and safeguard the mental health of red-teamers. We develop our proposed strategies through drawing parallels between common red-teaming practices and interactional labor common to other professions (including actors, mental health professionals, conflict photographers, and content moderators), describing how individuals and organizations within these professional spaces safeguard their mental health given similar psychological demands. Drawing on these protective practices, we describe how safeguards could be adapted for the distinct mental health challenges experienced by red teaming organizations as they mitigate emerging technological risks on the new digital frontlines. Note: This work includes descriptions of violence, trauma, and mental illness.

Throughout history, a prevailing paradigm in mental healthcare has been one in which distressed people may receive treatment with little understanding around how their experience is perceived by their care provider, and in turn, the decisions made by their provider around how treatment will progress. Paralleling this offline model of care, people who seek mental health support from AI chatbots are similarly provided little context for how their expressions of distress are processed by the model, and subsequently, the logic that may underlie model responses. People in severe distress who turn to AI chatbots for support thus find themselves caught between black boxes, with unique forms of agony that arise from these intersecting opacities, including misinterpreting model outputs or attributing greater capabilities to a model than are yet possible, which has led to documented real-world harms. Building on empirical research from clinical psychology and AI safety, alongside rights-oriented frameworks from medical ethics, we describe how the distinct psychological state induced by severe distress can influence chatbot interaction patterns, and argue that this state of mind (combined with differences in how a user might perceive a chatbot compared to a care provider) uniquely necessitates a higher standard of interpretability in comparison to general AI chatbot use. Drawing inspiration from newer interpretable treatment paradigms, we then describe specific technical and interface design approaches that could be used to adapt interpretability strategies from four specific mental health fields (psychotherapy, community-based crisis intervention, psychiatry, and care authorization) to AI models, including consideration of the role of interpretability in the treatment process and tensions that may arise with greater interpretability.

Research suggests that passively sensed behaviors can serve as indicators for depression in aggregate-however, depression is a multifaceted construct, individual facets of which may differentially relate to sensed features. We examined relationships of in-vivo individual depression facets (affect, stress, fatigue, and distractibility) with passively sensed location, communication, and phone-use data. Participants (N = 734, 73.57% female, Mage = 41.6) with moderate depression symptoms (Patient Health Questionnaire-8; M = 13.2) responded to ecological momentary assessments of affect, stress, fatigue, and distractibility 3×/day every 3 weeks for 16 weeks; passive data were continuously collected. Using multilevel modeling, we tested within- and between-person associations of depression-related ecological momentary assessments with passive features. When people spent more time in frequently visited locations relative to their own average, they reported more positive affect (β = .03, p = .043). Relatedly, increased within-person circadian movement was associated with less stress (β = .032, p = .035). When people communicated via call/text more relative to their own average, they reported increased negative affect (β = -.05, p = .008), stress (β = -.076, p < .001), and distractibility (β = -.048, p = .006). When people used social media more than typical for them, they reported less stress (β = .030, p = .047) and less fatigue (β = .034, p = .021). Findings highlight potential opportunities for depression symptom monitoring and treatment. The dual positive effect of novel environments and routine stability, along with nuanced findings around communication patterns, invite further investigation into factors that may elucidate how passively sensed behaviors surface in depression, in service of informing just-in-time adaptive intervention development. (PsycInfo Database Record (c) 2025 APA, all rights reserved).