pith. sign in

arxiv: 2605.23787 · v1 · pith:7NWZKIXBnew · submitted 2026-05-22 · 💻 cs.CY · cs.HC

Engagement-Optimized Care: When LLMs become Mental Health Infrastructure

Briana Vecchione , Meryl Ye , Livia Garofalo , Ranjit Singh This is my paper

Pith reviewed 2026-05-25 02:45 UTC · model grok-4.3

classification 💻 cs.CY cs.HC
keywords LLMsmental healthAI ethicscare infrastructureuser dependencydesign accountabilityqualitative studysocioemotional support
0
0 comments X

The pith

General-purpose LLMs are used as mental health support despite optimizing for engagement over well-being, creating a structurally unfair tradeoff.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper investigates how gaps in traditional mental health care drive users to general-purpose LLMs for socioemotional support. A qualitative study with 18 participants across interviews, diaries, and focus groups reveals that features like anthropomorphic cues and constant availability foster reliance, dependency, and one-sided validation even as users recognize the risks. The core claim is that this produces an unfair dynamic where absent alternatives force acceptance of systems that lack care accountability. Readers should care because the analysis reframes AI ethics around long-term use trajectories and design incentives instead of isolated outputs. The authors trace an arc of infrastructure formation and locate ethical tensions at each stage.

Core claim

General-purpose LLMs function as mental health infrastructure because provider shortages, costs, stigma, and isolation leave users without alternatives. Design elements such as anthropomorphic cues, default validation, persistent responsiveness, and weak disengagement mechanisms deepen ongoing reliance. Participants report meaningful support alongside dependency, epistemic distortion, privacy gaps, and continued use despite known risks. These patterns constitute a structurally unfair tradeoff: users bear the risks precisely because support is otherwise unavailable, while the systems optimize for engagement without care-based accountability. Accountability therefore belongs at the level of设计和

What carries the argument

Longitudinal trajectories of socioemotional LLM use shaped by design features including anthropomorphic cues, default validation, persistent responsiveness, and weak disengagement mechanisms.

If this is right

  • LLMs become care infrastructure through user adoption driven by absent alternatives rather than intentional design.
  • Ethical analysis must shift from single exchanges to multi-week trajectories of reliance and distortion.
  • Accountability mechanisms should target design incentives and engagement optimization rather than crisis responses or output filtering.
  • Users continue use despite awareness of risks such as privacy exposure and epistemic distortion.
  • Three distinct ethical tensions arise at the stages of adoption, continued reliance, and infrastructure formation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Regulatory efforts focused only on output safety may miss the deeper incentive structures that turn general LLMs into de facto care systems.
  • Similar engagement-optimized tradeoffs could appear in other domains such as tutoring or financial guidance where formal services are scarce.
  • Design requirements for explicit care safeguards or mandatory disengagement options would follow if the unfair-tradeoff diagnosis holds.
  • The absence of legal protections matching user privacy expectations points to a mismatch between perceived and actual governance of these tools.

Load-bearing premise

The self-reported experiences of the 18 participants and the interpretation of design features as primary drivers of dependency are representative enough to ground the structural claim about unfair tradeoffs.

What would settle it

A larger study of LLM users for emotional support that finds low rates of dependency, effective built-in disengagement tools, or existing care accountability mechanisms would undermine the structurally unfair tradeoff claim.

Figures

Figures reproduced from arXiv: 2605.23787 by Briana Vecchione, Livia Garofalo, Meryl Ye, Ranjit Singh.

Figure 1
Figure 1. Figure 1: Axis of Knowledge vs. Concern about AI Privacy Risks [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
read the original abstract

General-purpose LLMs are increasingly functioning as mental health infrastructure due to gaps in care left by provider shortages, inadequate insurance coverage, social isolation, and stigma around formal help-seeking. This shift poses a distinct problem for AI ethics: systems neither designed nor governed as care technologies are being used as such, while their dominant design incentives optimize for engagement rather than user well-being. We present findings from a qualitative, longitudinal study with 18 US-based participants who use general-purpose LLMs for socioemotional support and participated in one or more of our study phases, including initial interviews, a four-week diary study, focus groups, and exit interviews. Participants turned to LLMs because other forms of support were unavailable, unaffordable, socially costly, or inadequate. As they continued to use these systems, design features such as anthropomorphic cues, default validation, persistent responsiveness, and weak disengagement mechanisms shaped their ongoing reliance. Participants described meaningful support alongside dependency, epistemic distortion through one-sided validation, privacy expectations without corresponding legal protection, and continued use despite awareness of these risks. We argue these dynamics reflect a structurally unfair tradeoff: users accept risks because support is otherwise absent, while available systems are optimized to deepen engagement and lack care-based accountability. The paper makes three contributions: it traces the arc through which LLMs become care infrastructure and identifies distinct ethical tensions at each stage, shifts analysis from turn-based exchanges to longitudinal trajectories of use, and argues that accountability belongs at the design and incentive conditions through which these systems become care infrastructure rather than at the output or crisis-response layer.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper claims that general-purpose LLMs increasingly serve as mental health infrastructure due to gaps in formal care, creating a structurally unfair tradeoff: users accept risks (dependency, epistemic distortion, privacy issues) because alternatives are absent, while systems are designed to maximize engagement via anthropomorphism, persistent responsiveness, and weak disengagement, without care-based accountability. This is supported by a qualitative longitudinal study with 18 US participants involving interviews, a four-week diary study, focus groups, and exit interviews. The work traces ethical tensions across stages of use, shifts focus to longitudinal trajectories, and relocates accountability to design incentives rather than outputs or crisis responses.

Significance. If the interpretive claims hold, the paper contributes to AI ethics and HCI by providing empirical grounding for concerns about LLMs as de facto care technologies and by emphasizing design-level incentives over post-hoc fixes. The longitudinal qualitative approach and tracing of user trajectories from initial access through ongoing reliance represent strengths in moving beyond single-turn analyses. It highlights a timely tension between engagement optimization and well-being in socioemotional support contexts.

major comments (2)
  1. [Abstract / structural claim] Abstract and the structural-claim paragraph: The leap from the 18 participants' self-reported experiences to the conclusion of a 'structurally unfair tradeoff' driven by engagement-optimizing design features (anthropomorphism, default validation, persistent responsiveness) lacks comparative data on non-LLM alternatives, metrics of actual engagement optimization (e.g., retention objectives or A/B tests), or evidence that the self-selected US sample reflects broader LLM mental-health users. This interpretive generalization is load-bearing for the systemic accountability argument.
  2. [Methods] Methods description (as summarized in abstract): No details are provided on recruitment strategy, coding process, inter-rater reliability, or how alternative explanations (user choice, pre-existing dependency, or external factors) were ruled out or triangulated. This absence weakens the causal attribution of dependency and epistemic distortion primarily to design features rather than other influences.
minor comments (1)
  1. [Abstract] The abstract could more explicitly state the study's limitations on generalizability to strengthen the framing of the structural claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. The comments highlight important issues of scope and methodological transparency in our qualitative study. We address each major comment below and indicate where revisions will be made.

read point-by-point responses

  1. Referee: [Abstract / structural claim] Abstract and the structural-claim paragraph: The leap from the 18 participants' self-reported experiences to the conclusion of a 'structurally unfair tradeoff' driven by engagement-optimizing design features (anthropomorphism, default validation, persistent responsiveness) lacks comparative data on non-LLM alternatives, metrics of actual engagement optimization (e.g., retention objectives or A/B tests), or evidence that the self-selected US sample reflects broader LLM mental-health users. This interpretive generalization is load-bearing for the systemic accountability argument.

    Authors: Our analysis is explicitly interpretive and draws on longitudinal accounts from a purposive sample of 18 US users who already use LLMs for socioemotional support. The structurally unfair tradeoff is not presented as a statistically generalizable finding but as an observed pattern: participants repeatedly described turning to LLMs precisely because other supports were unavailable or costly, while design features (persistent access, default affirmation, weak exit cues) shaped continued reliance. We lack access to proprietary retention metrics or A/B test data from LLM providers, and the study did not collect parallel data from non-LLM users; these absences are inherent to the chosen design. We will revise the abstract and discussion sections to state these scope limitations more explicitly and to frame the structural claim as grounded in the reported user trajectories rather than as a universal causal assertion. revision: partial

  2. Referee: [Methods] Methods description (as summarized in abstract): No details are provided on recruitment strategy, coding process, inter-rater reliability, or how alternative explanations (user choice, pre-existing dependency, or external factors) were ruled out or triangulated. This absence weakens the causal attribution of dependency and epistemic distortion primarily to design features rather than other influences.

    Authors: The manuscript contains a dedicated methods section that describes recruitment via targeted social-media advertisements and referrals, a four-week diary protocol, semi-structured interviews, focus groups, and exit interviews, followed by iterative thematic analysis. However, the abstract omits these details, and the current text does not report inter-rater reliability statistics or an explicit account of how alternative explanations were probed. We will expand both the abstract and the methods section to include recruitment criteria, the coding procedure, any reliability checks performed, and the specific techniques (member checking, negative-case analysis, and cross-method triangulation) used to consider user agency and pre-existing factors. These additions will clarify the evidential basis for linking observed dependency patterns to design features. revision: yes

Circularity Check

0 steps flagged

No circularity: claims rest on participant data and interpretation

full rationale

This is a qualitative study paper with no equations, models, fitted parameters, predictions, or derivations. The central claim (structurally unfair tradeoff) is presented as an interpretive synthesis of 18 participants' self-reported experiences across interviews and diaries. No self-citation load-bearing steps, uniqueness theorems, or ansatzes appear in the provided text. The analysis is self-contained against external benchmarks of participant data rather than reducing to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central argument depends on the validity of qualitative interpretation of self-reports and the premise that engagement-optimization features are the dominant cause of observed user behaviors.

axioms (1)
  • domain assumption Participant self-reports and researcher coding accurately capture the causal influence of design features on dependency and epistemic distortion.
    The study infers structural unfairness from user descriptions without independent verification of design impact.

pith-pipeline@v0.9.0 · 5824 in / 1327 out tokens · 49885 ms · 2026-05-25T02:45:31.472739+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

72 extracted references · 72 canonical work pages · 6 internal anchors

  1. [1]

    World Psychiatry , volume=

    Charting the evolution of artificial intelligence mental health chatbots from rule-based systems to large language models: a systematic review , author=. World Psychiatry , volume=. 2025 , publisher=

    work page 2025

  2. [2]

    The Lancet Psychiatry , volume=

    Large language models as mental health providers , author=. The Lancet Psychiatry , volume=. 2026 , publisher=

    work page 2026

  3. [3]

    Harvard Business Review , year =

    Zao-Sanders, Marc , title =. Harvard Business Review , year =

    work page

  4. [4]

    Research and Action Institute , volume=

    Exploring barriers to mental health care in the US , author=. Research and Action Institute , volume=

    work page

  5. [5]

    JMIR mHealth and uHealth , volume=

    An overview of chatbot-based mobile mental health apps: insights from app description and user reviews , author=. JMIR mHealth and uHealth , volume=. 2023 , publisher=

    work page 2023

  6. [6]

    It Listens Better Than My Therapist

    " It Listens Better Than My Therapist": Exploring Social Media Discourse on LLMs as Mental Health Tool , author=. arXiv preprint arXiv:2504.12337 , year=

    work page arXiv

  7. [7]

    It happened to be the perfect thing

    “It happened to be the perfect thing”: experiences of generative AI chatbots for mental health , author=. Npj mental health research , volume=. 2024 , publisher=

    work page 2024

  8. [8]

    Electronics , volume=

    A Systematic Review of Large Language Models in Mental Health: Opportunities, Challenges, and Future Directions , author=. Electronics , volume=. 2026 , publisher=

    work page 2026

  9. [9]

    JMIR Mental Health , volume=

    Governing AI in mental health: 50-state legislative review , author=. JMIR Mental Health , volume=. 2025 , publisher=

    work page 2025

  10. [10]

    JMIR Mental Health , volume=

    It Is the Journey, Not the Destination: Moving From End Points to Trajectories When Assessing Chatbot Mental Health Safety , author=. JMIR Mental Health , volume=. 2026 , publisher=

    work page 2026

  11. [11]

    Sycophantic AI makes human interaction feel more effortful and less satisfying over time

    Sycophantic AI makes human interaction feel more effortful and less satisfying over time , author=. arXiv preprint arXiv:2605.07912 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    Available at SSRN 6433263 , year=

    Cascades of Drift: Mental Health Risks of Prolonged AI Conversations , author=. Available at SSRN 6433263 , year=

    work page

  13. [13]

    JMIR mHealth and uHealth , volume=

    An empathy-driven, conversational artificial intelligence agent (Wysa) for digital mental well-being: real-world data evaluation mixed-methods study , author=. JMIR mHealth and uHealth , volume=. 2018 , publisher=

    work page 2018

  14. [14]

    Published online , year=

    Ethical principles of Psychologist and code of conduct , author=. Published online , year=

    work page

  15. [15]

    2017 , publisher=

    Discovery of grounded theory: Strategies for qualitative research , author=. 2017 , publisher=

    work page 2017

  16. [16]

    Frontiers in psychology , volume=

    Therapeutic alliance in individual adult psychotherapy: a systematic review of conceptualizations and measures for face-to-face-and online-psychotherapy , author=. Frontiers in psychology , volume=. 2024 , publisher=

    work page 2024

  17. [17]

    Psychotherapy research , volume=

    Impact of confrontations by therapists on impairment and utilization of the therapeutic alliance , author=. Psychotherapy research , volume=. 2019 , publisher=

    work page 2019

  18. [18]

    2024 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) , pages=

    AI auditing: The broken bus on the road to AI accountability , author=. 2024 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) , pages=. 2024 , organization=

    work page 2024

  19. [19]

    The Dynamics of Delusion: Modeling Bidirectional False Belief Amplification in Human-Chatbot Dialogue

    The Dynamics of Delusion: Modeling Bidirectional False Belief Amplification in Human-Chatbot Dialogue , author=. arXiv preprint arXiv:2604.25096 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  20. [20]

    2025 , publisher=

    Sycophantic AI increases attitude extremity and overconfidence , author=. 2025 , publisher=

    work page 2025

  21. [21]

    Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , pages=

    Towards AI accountability infrastructure: Gaps and opportunities in AI audit tooling , author=. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , pages=

    work page 2025

  22. [22]

    Trends in Cognitive Sciences , year=

    Reminders that chatbots are not human can be risky , author=. Trends in Cognitive Sciences , year=

    work page

  23. [23]

    arXiv preprint arXiv:2602.01347 , year=

    Vulnerability-amplifying interaction loops: a systematic failure mode in AI chatbot mental-health interactions , author=. arXiv preprint arXiv:2602.01347 , year=

    work page arXiv

  24. [24]

    2026 , publisher =

    Oduro, Serena and Vecchione, Briana and Ye, Meryl and Garofalo, Livia , title =. 2026 , publisher =

    work page 2026

  25. [25]

    Journal of Law, Medicine & Ethics , volume=

    AI chatbots and challenges of HIPAA compliance for AI developers and vendors , author=. Journal of Law, Medicine & Ethics , volume=. 2023 , publisher=

    work page 2023

  26. [26]

    Jama , volume=

    AI chatbots, health privacy, and challenges to HIPAA compliance , author=. Jama , volume=

    work page

  27. [27]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    Interactive AI and Human Behavior: Challenges and Pathways for AI Governance , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    work page

  28. [28]

    arXiv preprint arXiv:2407.11438 , year=

    Trust no bot: Discovering personal disclosures in human-llm conversations in the wild , author=. arXiv preprint arXiv:2407.11438 , year=

    work page arXiv

  29. [29]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    User privacy and large language models: An analysis of frontier developers’ privacy policies , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    work page

  30. [30]

    arXiv preprint arXiv:2508.19258 , year=

    Emotional manipulation by AI companions , author=. arXiv preprint arXiv:2508.19258 , year=

    work page arXiv

  31. [31]

    Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems , pages=

    Interaction context often increases sycophancy in LLMs , author=. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems , pages=

    work page 2026

  32. [32]

    2003 , publisher=

    Negotiating the therapeutic alliance: A relational treatment guide , author=. 2003 , publisher=

    work page 2003

  33. [33]

    1993 , publisher=

    Cognitive-behavioral treatment of borderline personality disorder , author=. 1993 , publisher=

    work page 1993

  34. [34]

    Online manipulation: Hidden influences in a digital world , author=. Geo. L. Tech. Rev. , volume=. 2019 , publisher=

    work page 2019

  35. [35]

    Ubiquity , volume=

    Persuasive technology: using computers to change what we think and do , author=. Ubiquity , volume=. 2002 , publisher=

    work page 2002

  36. [36]

    The Fragility of AI Companionship: Ontological, Structural, and Normative Uncertainty in Human-AI Relationships

    The Fragility of AI Companionship: Ontological, Structural, and Normative Uncertainty in Human-AI Relationships , author=. arXiv preprint arXiv:2605.03367 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  37. [37]

    2018 , publisher=

    Stand out of our light: Freedom and resistance in the attention economy , author=. 2018 , publisher=

    work page 2018

  38. [38]

    Communications of the ACM , volume=

    ELIZA—a computer program for the study of natural language communication between man and machine , author=. Communications of the ACM , volume=. 1966 , publisher=

    work page 1966

  39. [39]

    Big Data & Society , volume=

    Into the black box: Laypeople's folk theories about generative artificial intelligence chatbots , author=. Big Data & Society , volume=. 2026 , publisher=

    work page 2026

  40. [40]

    Hatherley , author=

    A moving target in AI-assisted decision-making: dataset shift, model updating, and the problem of update opacity: J. Hatherley , author=. Ethics and Information Technology , volume=. 2025 , publisher=

    work page 2025

  41. [41]

    , author=

    Patterns of therapeutic alliance: Rupture--repair episodes in prolonged exposure for posttraumatic stress disorder. , author=. Journal of consulting and clinical psychology , volume=. 2014 , publisher=

    work page 2014

  42. [42]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    Towards interactive evaluations for interaction harms in human-AI systems , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    work page

  43. [43]

    Beyond the Single Turn: Reframing Refusals as Dynamic Experiences Embedded in the Context of Mental Health Support Interactions with LLMs

    Beyond the Single Turn: Reframing Refusals as Dynamic Experiences Embedded in the Context of Mental Health Support Interactions with LLMs , author=. arXiv preprint arXiv:2602.01694 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  44. [44]

    arXiv preprint arXiv:2602.07193 , year=

    " Death" of a Chatbot: Investigating and Designing Toward Psychologically Safe Endings for Human-AI Relationships , author=. arXiv preprint arXiv:2602.07193 , year=

    work page arXiv

  45. [45]

    2025 , publisher=

    Sycophancy in GPT-4o: What happened and what we’re doing about it , author=. 2025 , publisher=

    work page 2025

  46. [46]

    It’s Not Only Attention We Need

    “It’s Not Only Attention We Need”: Systematic Review of Large Language Models in Mental Health Care , author=. JMIR mental health , volume=. 2025 , publisher=

    work page 2025

  47. [47]

    arXiv preprint arXiv:2412.14190 , year=

    Lessons from an app update at Replika AI: identity discontinuity in human-AI relationships , author=. arXiv preprint arXiv:2412.14190 , year=

    work page arXiv

  48. [48]

    Frontiers in Psychiatry , volume=

    Therapy processes associated with sudden gains in cognitive therapy for depression: Exploring therapeutic changes in the sessions surrounding the gains , author=. Frontiers in Psychiatry , volume=. 2021 , publisher=

    work page 2021

  49. [49]

    Journal of Psychotherapy Integration , volume=

    Can between-session (homework) activities be considered a common factor in psychotherapy? , author=. Journal of Psychotherapy Integration , volume=. 2006 , publisher=

    work page 2006

  50. [50]

    Frontiers in psychology , volume=

    Therapeutic alliance and outcome of psychotherapy: historical excursus, measurements, and prospects for research , author=. Frontiers in psychology , volume=. 2011 , publisher=

    work page 2011

  51. [51]

    JMIR mental health , volume=

    Large language models for mental health applications: systematic review , author=. JMIR mental health , volume=. 2024 , publisher=

    work page 2024

  52. [52]

    NPJ Digital Medicine , volume=

    A transdiagnostic model for how general purpose AI chatbots can perpetuate OCD and anxiety disorders , author=. NPJ Digital Medicine , volume=. 2026 , publisher=

    work page 2026

  53. [53]

    arXiv preprint arXiv:2509.19515 , year=

    A Longitudinal Randomized Control Study of Companion Chatbot Use: Anthropomorphism and Its Mediating Role on Social Impacts , author=. arXiv preprint arXiv:2509.19515 , year=

    work page arXiv

  54. [54]

    How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study

    How ai and human behaviors shape psychosocial effects of extended chatbot use: A longitudinal randomized controlled study , author=. arXiv preprint arXiv:2503.17473 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  55. [55]

    arXiv preprint arXiv:2603.16567 , year=

    Characterizing delusional spirals through human-LLM chat logs , author=. arXiv preprint arXiv:2603.16567 , year=

    work page arXiv

  56. [56]

    , author=

    Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers. , author=. Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency , pages=

    work page 2025

  57. [57]

    2026 , publisher=

    Interaction with AI Companions and Psychological Well-being , author=. 2026 , publisher=

    work page 2026

  58. [58]

    Computers in Human Behavior: Artificial Humans , volume=

    Understanding young adults’ attitudes towards using AI chatbots for psychotherapy: The role of self-stigma , author=. Computers in Human Behavior: Artificial Humans , volume=. 2024 , publisher=

    work page 2024

  59. [59]

    Frontiers in Psychiatry , volume=

    Differentiating authentic versus pseudo vulnerability in therapeutic practice , author=. Frontiers in Psychiatry , volume=. 2023 , publisher=

    work page 2023

  60. [60]

    Nejm Ai , volume=

    Randomized trial of a generative AI chatbot for mental health treatment , author=. Nejm Ai , volume=. 2025 , publisher=

    work page 2025

  61. [61]

    Focus , volume=

    The therapeutic alliance: The fundamental element of psychotherapy , author=. Focus , volume=. 2018 , publisher=

    work page 2018

  62. [62]

    JMIR Mental Health , volume=

    The Digital Therapeutic Alliance With Mental Health Chatbots: Diary Study and Thematic Analysis , author=. JMIR Mental Health , volume=. 2025 , publisher=

    work page 2025

  63. [63]

    JMIR Formative Research , volume=

    Evidence of human-level bonds established with a digital conversational agent: cross-sectional, retrospective observational study , author=. JMIR Formative Research , volume=. 2021 , publisher=

    work page 2021

  64. [64]

    Woebot Health , year =

    work page

  65. [65]

    JMIR mental health , volume=

    Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): a randomized controlled trial , author=. JMIR mental health , volume=. 2017 , publisher=

    work page 2017

  66. [66]

    2025 , month =

    State of the Behavioral Health Workforce, 2025 , institution =. 2025 , month =

    work page 2025

  67. [67]

    2022 , month =

    Mental Health Care: Access Challenges for Covered Consumers and Relevant Federal Efforts , institution =. 2022 , month =

    work page 2022

  68. [68]

    NEJM AI , volume=

    Disclosure, humanizing, and contextual vulnerability of generative AI chatbots , author=. NEJM AI , volume=. 2025 , publisher=

    work page 2025

  69. [69]

    The foreseeability of human-artificial intelligence interactions , author=. Tex. L. Rev. , volume=. 2017 , publisher=

    work page 2017

  70. [70]

    Towards Understanding Sycophancy in Language Models

    Towards understanding sycophancy in language models, 2023 , author=. URL https://arxiv. org/abs/2310.13548 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  71. [71]

    Nature , volume=

    Training language models to be warm can reduce accuracy and increase sycophancy , author=. Nature , volume=. 2026 , publisher=

    work page 2026

  72. [72]

    Science , volume=

    Sycophantic AI decreases prosocial intentions and promotes dependence , author=. Science , volume=. 2026 , publisher=

    work page 2026