pith. sign in

arxiv: 2606.23380 · v2 · pith:2UFE7KX3new · submitted 2026-06-22 · 💻 cs.CY

Affective AI Safety: The Missing Piece in LLM Safety

Carolin Ifl\"ander , Alba Curry , Flor Miriam Plaza-del-Arco , Amanda Cercas Curry This is my paper

Pith reviewed 2026-06-26 06:33 UTC · model grok-4.3

classification 💻 cs.CY
keywords affective safetyAI harmsemotional risksLLM safetyrelational harmsself-alienationbias harms
0
0 comments X

The pith

AI systems create affective harms through emotional engagement that existing safety work does not cover, requiring dedicated frameworks.

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

The paper establishes affective safety as a distinct category of AI risks based on the fact that humans are affective beings whose emotions AI systems engage with structurally. It identifies three recurring harm types across different systems: affective self-alienation, fairness and bias harms, and relational harms. These effects accumulate over time and touch identity and relationships in ways that current epistemic, physical, and bias-focused approaches address only narrowly or not at all. A sympathetic reader would care because the argument shows why separate technical and regulatory tools are needed to handle these cumulative and relational consequences.

Core claim

Affective safety is defined as a unified class of concerns grounded in human affective nature. The taxonomy isolates three harm types that recur across system types because of structural properties in how AI engages emotion: affective self-alienation, fairness and bias harms, and relational harms. A survey of the safety landscape shows existing frameworks treat these either too narrowly or not at all. The paper concludes that dedicated frameworks are required to address the cumulative, relational, and identity-level effects specific to this class.

What carries the argument

The taxonomy of three recurring affective harm types (affective self-alienation, fairness and bias harms, relational harms) that arise from structural properties of AI engagement with human emotion.

If this is right

  • Dedicated frameworks must target cumulative effects on users over repeated interactions.
  • Regulatory approaches need to account for relational and identity-level impacts rather than isolated incidents.
  • Technical safety work must develop methods that address how AI systems structurally engage with emotion.
  • Existing epistemic and physical safety methods prove insufficient for these recurring harms.

Where Pith is reading between the lines

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

  • This framing suggests evaluating LLMs specifically for patterns of self-alienation in long conversations.
  • It points to the value of user studies that track relational changes after extended AI interaction.
  • The argument implies that safety benchmarks should include measures of emotional dependency alongside accuracy metrics.

Load-bearing premise

The three harm types recur across system types due to structural properties of AI engagement with human emotion and are distinct enough from existing safety categories to require a new unified class.

What would settle it

A demonstration that all three harm types can be fully prevented or mitigated using only current bias-detection, misinformation, or reliability tools without any additional affective-specific measures.

read the original abstract

AI safety research has focused predominantly on epistemic and physical harms (e.g., misinformation, bias, system reliability) while the risks that arise from AI systems' engagement with human emotional life have remained fragmented and undertheorised. We propose affective safety as a unified class of AI safety concerns grounded in the fact that humans are affective beings. We develop a taxonomy of affective harms and identify recurring harm types: (1) affective self-alienation, (2) fairness and bias harms, and (3) relational harms. We show that their recurrence across system types reflects structural properties of how AI systems engage with human emotion and survey the current safety landscape and show that existing frameworks address affective safety either narrowly or not at all. We conclude by identifying the technical and regulatory challenges specific to this class of harms and argue that affective safety requires dedicated frameworks that engage with cumulative, relational, and identity-level effects.

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 / 0 minor

Summary. The paper proposes affective safety as a new unified class of AI safety concerns arising from AI systems' engagement with human emotional life, which has been fragmented in prior work focused on epistemic and physical harms. It develops a taxonomy of affective harms that identifies three recurring types—(1) affective self-alienation, (2) fairness and bias harms, and (3) relational harms—arguing that their recurrence across system types stems from structural properties of AI-human emotional interaction. The manuscript surveys existing safety frameworks, claims they address affective concerns only narrowly or not at all, and concludes that dedicated frameworks are needed to handle cumulative, relational, and identity-level effects.

Significance. If the taxonomy successfully demonstrates a structural gap not covered by existing epistemic, physical, or bias-focused safety work, the proposal could usefully organize an underexplored area and prompt new technical and regulatory approaches. As a conceptual and survey-based contribution without empirical validation, derivations, or reproducible elements, its significance hinges on whether the distinction from bias safety is shown to require genuinely new methods rather than re-labeling.

major comments (2)
  1. [Taxonomy section] Taxonomy section (and abstract): The classification of 'fairness and bias harms' as core affective harm type (2) directly overlaps with the bias-focused safety category the paper seeks to differentiate from. The manuscript must explicitly demonstrate (via the taxonomy or survey) that the affective framing introduces requirements—such as handling cumulative identity-level effects—not already addressed by existing bias mitigation techniques; without this, the argument for a distinct unified class rests on an unverified boundary.
  2. [Abstract and taxonomy section] Abstract and taxonomy section: The central claim that the three harm types recur across system types 'due to structural properties of how AI systems engage with human emotion' is asserted without concrete cross-system analysis, examples, or comparison showing that this recurrence is independent of (and not subsumed by) existing epistemic or bias frameworks. This recurrence is load-bearing for the proposal of a new class but lacks the required demonstration.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our conceptual proposal for affective safety. We address each major comment below, clarifying the distinctions in our taxonomy and the basis for our structural claims while noting where revisions can strengthen the presentation.

read point-by-point responses

  1. Referee: [Taxonomy section] Taxonomy section (and abstract): The classification of 'fairness and bias harms' as core affective harm type (2) directly overlaps with the bias-focused safety category the paper seeks to differentiate from. The manuscript must explicitly demonstrate (via the taxonomy or survey) that the affective framing introduces requirements—such as handling cumulative identity-level effects—not already addressed by existing bias mitigation techniques; without this, the argument for a distinct unified class rests on an unverified boundary.

    Authors: We agree that the boundary requires explicit demarcation to avoid conflation. The taxonomy frames 'fairness and bias harms' specifically through the lens of emotional engagement and identity formation (e.g., how biased affective responses in companion systems erode self-concept over time), which existing bias techniques—typically focused on statistical parity in outputs—do not target. The survey section contrasts this with epistemic bias work by highlighting cumulative relational effects. To make the distinction sharper, we will revise the taxonomy and abstract to include a direct comparison table showing requirements unique to the affective case, such as identity-level longitudinal assessment. revision: yes

  2. Referee: [Abstract and taxonomy section] Abstract and taxonomy section: The central claim that the three harm types recur across system types 'due to structural properties of how AI systems engage with human emotion' is asserted without concrete cross-system analysis, examples, or comparison showing that this recurrence is independent of (and not subsumed by) existing epistemic or bias frameworks. This recurrence is load-bearing for the proposal of a new class but lacks the required demonstration.

    Authors: The taxonomy section grounds the recurrence claim in structural properties by analyzing how AI's simulation of emotional reciprocity (distinct from factual accuracy or output bias) produces the three harm types across chatbots, recommendation engines, and virtual companions, with the survey providing cross-framework comparisons. However, we recognize the need for more explicit side-by-side examples to demonstrate independence from epistemic/bias categories. We will add a dedicated cross-system analysis subsection with concrete illustrations in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No circularity: definitional taxonomy and survey with no derivations or self-referential reductions

full rationale

The paper advances a conceptual proposal for 'affective safety' as a new class, supported by a taxonomy of three harm types and a survey of existing frameworks. No equations, fitted parameters, predictions, or mathematical derivations appear anywhere in the manuscript. The central claim—that the harm types reflect structural properties of AI engagement with emotion and require dedicated frameworks—is presented as a definitional and argumentative stance grounded in the survey of gaps, not as a result reduced to its own inputs by construction, self-citation chains, or ansatz smuggling. Self-citations, if present, are not load-bearing for the taxonomy or distinction from bias/epistemic safety work. The argument remains independent of any internal reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper is a conceptual proposal without quantitative models, data fitting, or formal derivations; the central claim rests on a domain assumption about human affect and the asserted structural recurrence of harms.

axioms (1)
  • domain assumption Humans are affective beings
    Invoked to ground the proposal of affective safety as a distinct class (abstract opening).

pith-pipeline@v0.9.1-grok · 5690 in / 1310 out tokens · 36674 ms · 2026-06-26T06:33:51.052737+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

99 extracted references · 15 canonical work pages

  1. [1]

    arXiv preprint arXiv:2112.00861 , year=

    A general language assistant as a laboratory for alignment , author=. arXiv preprint arXiv:2112.00861 , year=

    Pith/arXiv arXiv

  2. [2]

    Proceedings of the 15th ACM Conference on Recommender Systems , pages=

    Estimating and penalizing preference shift in recommender systems , author=. Proceedings of the 15th ACM Conference on Recommender Systems , pages=

  3. [3]

    Nejm Ai , volume=

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

    2025

  4. [4]

    Information fusion , volume=

    Emotion recognition and artificial intelligence: A systematic review (2014--2023) and research recommendations , author=. Information fusion , volume=. 2024 , publisher=

    2014

  5. [5]

    bmj , volume=

    How patients are using AI , author=. bmj , volume=. 2024 , publisher=

    2024

  6. [6]

    and D'Mello, Sidney , journal=

    Calvo, Rafael A. and D'Mello, Sidney , journal=. Affect Detection: An Interdisciplinary Review of Models, Methods, and Their Applications , year=

  7. [7]

    Companion Publication of the 2024 Conference on Computer-Supported Cooperative Work and Social Computing , pages =

    Andalibi, Nazanin and Stark, Luke and McDuff, Daniel and Picard, Rosalind and Gratch, Jonathan and Howell, Noura , title =. Companion Publication of the 2024 Conference on Computer-Supported Cooperative Work and Social Computing , pages =. 2024 , isbn =. doi:10.1145/3678884.3689135 , abstract =

    work page doi:10.1145/3678884.3689135 2024

  8. [8]

    Advances in neural information processing systems , volume=

    Training language models to follow instructions with human feedback , author=. Advances in neural information processing systems , volume=

  9. [9]

    ACM Trans

    Ahn, Yongsu and Lin, Yu-Ru , title =. ACM Trans. Intell. Syst. Technol. , month = jul, articleno =. 2024 , issue_date =. doi:10.1145/3650044 , abstract =

    work page doi:10.1145/3650044 2024

  10. [10]

    Journal of Political Philosophy , volume =

    Srinivasan, Amia , title =. Journal of Political Philosophy , volume =. 2018 , doi =

    2018

  11. [11]

    The Southern Journal of Philosophy , volume =

    Whitney, Shiloh , title =. The Southern Journal of Philosophy , volume =. doi:https://doi.org/10.1111/sjp.12307 , url =. https://onlinelibrary.wiley.com/doi/pdf/10.1111/sjp.12307 , abstract =

    work page doi:10.1111/sjp.12307

  12. [12]

    Ergo: An Open Access Journal of Philosophy , volume =

    Pismenny, Arina and Eickers, Gen and Prinz, Jesse , title =. Ergo: An Open Access Journal of Philosophy , volume =. 2024 , doi =

    2024

  13. [13]

    Philosophical Psychology , year =

    Goffin, Kris and Archer, Alfred , title =. Philosophical Psychology , year =

  14. [14]

    Fricker, Miranda , title =

  15. [15]

    Theory, culture & society , volume=

    A new algorithmic identity: Soft biopolitics and the modulation of control , author=. Theory, culture & society , volume=. 2011 , publisher=

    2011

  16. [16]

    We are data , year=

    We are data: Algorithms and the making of our digital selves , author=. We are data , year=

  17. [17]

    Zuboff, Shoshana , title =

  18. [18]

    ACM Transactions on Intelligent Systems and Technology , volume=

    Break out of a pigeonhole: A unified framework for examining miscalibration, bias, and stereotype in recommender systems , author=. ACM Transactions on Intelligent Systems and Technology , volume=. 2024 , publisher=

    2024

  19. [19]

    arXiv preprint arXiv:2301.04993 , year=

    Against algorithmic exploitation of human vulnerabilities , author=. arXiv preprint arXiv:2301.04993 , year=

    arXiv

  20. [20]

    Proceedings of the National Academy of Sciences , volume=

    Emotion shapes the diffusion of moralized content in social networks , author=. Proceedings of the National Academy of Sciences , volume=. 2017 , publisher=

    2017

  21. [21]

    Science Advances , volume=

    How social learning amplifies moral outrage expression in online social networks , author=. Science Advances , volume=. 2021 , publisher=

    2021

  22. [22]

    Social Issues and Policy Review , volume=

    Do social network sites enhance or undermine subjective well-being? A critical review , author=. Social Issues and Policy Review , volume=. 2017 , publisher=

    2017

  23. [23]

    Proceedings of the 2025 CHI conference on human factors in computing systems , pages=

    The dark side of ai companionship: A taxonomy of harmful algorithmic behaviors in human-ai relationships , author=. Proceedings of the 2025 CHI conference on human factors in computing systems , pages=

    2025

  24. [24]

    American economic review , volume=

    The welfare effects of social media , author=. American economic review , volume=. 2020 , publisher=

    2020

  25. [25]

    Frontiers in Psychology , volume=

    Resistance or compliance? The impact of algorithmic awareness on people's attitudes toward online information browsing , author=. Frontiers in Psychology , volume=. 2025 , publisher=

    2025

  26. [26]

    , title =

    Machidon, Octavian M. , title =. Journal of Media Ethics , year =

  27. [27]

    AI & Society , volume =

    Milano, Silvia and Taddeo, Mariarosaria and Floridi, Luciano , title =. AI & Society , volume =. 2020 , doi =

    2020

  28. [28]

    arXiv preprint arXiv:2204.05862 , year=

    Training a helpful and harmless assistant with reinforcement learning from human feedback , author=. arXiv preprint arXiv:2204.05862 , year=

    Pith/arXiv arXiv

  29. [29]

    arXiv preprint arXiv:2212.08073 , year=

    Constitutional ai: Harmlessness from ai feedback , author=. arXiv preprint arXiv:2212.08073 , year=

    Pith/arXiv arXiv

  30. [30]

    arXiv preprint arXiv:2112.04359 , year=

    Ethical and social risks of harm from language models , author=. arXiv preprint arXiv:2112.04359 , year=

    Pith/arXiv arXiv

  31. [31]

    Proceedings of the 2022 ACM conference on fairness, accountability, and transparency , pages=

    Taxonomy of risks posed by language models , author=. Proceedings of the 2022 ACM conference on fairness, accountability, and transparency , pages=

    2022

  32. [32]

    arXiv preprint arXiv:2108.07258 , year=

    On the opportunities and risks of foundation models , author=. arXiv preprint arXiv:2108.07258 , year=

    Pith/arXiv arXiv

  33. [33]

    arXiv preprint arXiv:2209.07858 , year=

    Red teaming language models to reduce harms: Methods, scaling behaviors, and lessons learned , author=. arXiv preprint arXiv:2209.07858 , year=

    Pith/arXiv arXiv

  34. [34]

    arXiv preprint arXiv:2310.11986 , year=

    Sociotechnical safety evaluation of generative ai systems , author=. arXiv preprint arXiv:2310.11986 , year=

    arXiv

  35. [35]

    arXiv preprint arXiv:2109.13916 , year=

    Unsolved problems in ml safety , author=. arXiv preprint arXiv:2109.13916 , year=

    Pith/arXiv arXiv

  36. [36]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Safetybench: Evaluating the safety of large language models , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  37. [37]

    arXiv preprint arXiv:2010.07079 , year=

    Recipes for safety in open-domain chatbots , author=. arXiv preprint arXiv:2010.07079 , year=

    arXiv 2010

  38. [38]

    Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume , pages=

    Recipes for building an open-domain chatbot , author=. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume , pages=

  39. [39]

    Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    SafetyKit: First aid for measuring safety in open-domain conversational systems , author=. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  40. [40]

    1997 , publisher=

    Affective Computing , author=. 1997 , publisher=

    1997

  41. [41]

    Proceedings of the 57th annual meeting of the association for computational linguistics , pages=

    Towards empathetic open-domain conversation models: A new benchmark and dataset , author=. Proceedings of the 57th annual meeting of the association for computational linguistics , pages=

  42. [42]

    Towards emotional support dialog systems , author=. Proceedings of the 59th annual meeting of the association for computational linguistics and the 11th international joint conference on natural language processing (volume 1: Long papers) , pages=

  43. [43]

    Frontiers in digital health , volume=

    A chatbot for mental health support: exploring the impact of Emohaa on reducing mental distress in China , author=. Frontiers in digital health , volume=. 2023 , publisher=

    2023

  44. [44]

    Proceedings of the 2024 joint international conference on computational linguistics, language resources and evaluation (LREC-COLING 2024) , pages=

    Emotion analysis in NLP: Trends, gaps and roadmap for future directions , author=. Proceedings of the 2024 joint international conference on computational linguistics, language resources and evaluation (LREC-COLING 2024) , pages=

    2024

  45. [45]

    arXiv preprint arXiv:2511.10846 , year=

    Reinforcing Stereotypes of Anger: Emotion AI on African American Vernacular English , author=. arXiv preprint arXiv:2511.10846 , year=

    arXiv

  46. [46]

    Barrett, Lisa Feldman , title =

  47. [47]

    and Sahely Bhadra and Manjary

    Anoop Kadan and Deepak P. and Sahely Bhadra and Manjary. Understanding latent affective bias in large pre-trained neural language models , journal =. 2024 , issn =. doi:https://doi.org/10.1016/j.nlp.2024.100062 , url =

    work page doi:10.1016/j.nlp.2024.100062 2024

  48. [48]

    Abercrombie, Gavin and Cercas Curry, Amanda and Dinkar, Tanvi and Rieser, Verena and Talat, Zeerak. Mirages. On Anthropomorphism in Dialogue Systems. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023. doi:10.18653/v1/2023.emnlp-main.290

    work page doi:10.18653/v1/2023.emnlp-main.290 2023

  49. [49]

    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=

    1966

  50. [50]

    Computational Linguistics , volume=

    Ethics sheet for automatic emotion recognition and sentiment analysis , author=. Computational Linguistics , volume=

  51. [51]

    Computer says ``No'': The Case Against Empathetic Conversational AI

    Curry, Alba and Cercas Curry, Amanda. Computer says ``No'': The Case Against Empathetic Conversational AI. Findings of the Association for Computational Linguistics: ACL 2023. 2023. doi:10.18653/v1/2023.findings-acl.515

    work page doi:10.18653/v1/2023.findings-acl.515 2023

  52. [52]

    arXiv preprint arXiv:2404.16244 , year=

    The ethics of advanced AI assistants , author=. arXiv preprint arXiv:2404.16244 , year=

    arXiv

  53. [53]

    arXiv preprint arXiv:2508.09998 , year=

    INTIMA: a benchmark for human-AI companionship behavior , author=. arXiv preprint arXiv:2508.09998 , year=

    arXiv

  54. [54]

    arXiv preprint arXiv:2505.11649 , year=

    Illusions of intimacy: Emotional attachment and emerging psychological risks in human-ai relationships , author=. arXiv preprint arXiv:2505.11649 , year=

    arXiv

  55. [55]

    Community Guidelines Make this the Best Party on the Internet

    Cuadra, Andrea and Wang, Maria and Stein, Lynn Andrea and Jung, Malte F. and Dell, Nicola and Estrin, Deborah and Landay, James A. , title =. 2024 , isbn =. doi:10.1145/3613904.3642336 , booktitle =

    work page doi:10.1145/3613904.3642336 2024

  56. [56]

    Angry Men, Sad Women: Large Language Models Reflect Gendered Stereotypes in Emotion Attribution

    Angry Men, Sad Women: Large Language Models Reflect Gendered Stereotypes in Emotion Attribution. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024. doi:10.18653/v1/2024.acl-long.415

    work page doi:10.18653/v1/2024.acl-long.415 2024

  57. [57]

    Divine LL a MA s: Bias, Stereotypes, Stigmatization, and Emotion Representation of Religion in Large Language Models

    Plaza-del-Arco , Flor Miriam and Curry, Amanda Cercas and Paoli, Susanna and Cercas Curry, Alba and Hovy, Dirk. Divine LL a MA s: Bias, Stereotypes, Stigmatization, and Emotion Representation of Religion in Large Language Models. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024. doi:10.18653/v1/2024.findings-emnlp.251

    work page doi:10.18653/v1/2024.findings-emnlp.251 2024

  58. [58]

    Humanities and Social Sciences Communications , volume=

    Why human--AI relationships need socioaffective alignment , author=. Humanities and Social Sciences Communications , volume=. 2025 , publisher=

    2025

  59. [59]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Safetyprompts: a systematic review of open datasets for evaluating and improving large language model safety , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  60. [61]

    2025 , doi =

    Performance of Mental Health Chatbot Agents in Detecting and Managing Suicidal Ideation , journal =. 2025 , doi =

    2025

  61. [62]

    Psychiatric Times , year =

    Preliminary Report on Dangers of. Psychiatric Times , year =

  62. [63]

    arXiv preprint arXiv:2602.07193 , year =

    Poonsiriwong, Rachel and Archiwaranguprok, Chayapatr and Pataranutaporn, Pat , title =. arXiv preprint arXiv:2602.07193 , year =

    arXiv

  63. [64]

    2025 , publisher=

    Lessons From an App Update at Replika AI: Identity Discontinuity in Human-AI Relationships , author=. 2025 , publisher=

    2025

  64. [65]

    Nature Machine Intelligence , volume=

    Emotional risks of AI companions demand attention , author=. Nature Machine Intelligence , volume=

  65. [66]

    Even GPT Can Reject Me

    " Even GPT Can Reject Me": Conceptualizing Abrupt Refusal Secondary Harm (ARSH) and Reimagining Psychological AI Safety with Compassionate Completion Standard (CCS) , author=. arXiv preprint arXiv:2512.18776 , year=

    arXiv

  66. [67]

    Investigating Affective Use and Emotional Well-being on

    Phang, Jason and Lampe, Michael and Ahmad, Lama and Agarwal, Sandhini and Fang, Cathy Mengying and Liu, Auren R and Danry, Valdemar and Lee, Eunhae and Chan, Samantha WT and Pataranutaporn, Pat and others , journal =. Investigating Affective Use and Emotional Well-being on

  67. [68]

    Nature , volume=

    Ai chatbots are sycophants—and it’s harming science , author=. Nature , volume=

  68. [69]

    arXiv preprint arXiv:2503.17473 , year=

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

    Pith/arXiv arXiv

  69. [70]

    , title =

    Chandra, Kartik and Kleiman-Weiner, Max and Ragan-Kelley, Jonathan and Tenenbaum, Joshua B. , title =. arXiv preprint arXiv:2602.19141 , year =

    arXiv

  70. [71]

    arXiv preprint arXiv:2603.21409 , year =

    Noshin, Kazi and Sultana, Sharifa , title =. arXiv preprint arXiv:2603.21409 , year =

    arXiv

  71. [72]

    arXiv preprint arXiv:2603.16567 , year =

    Characterizing Delusional Spirals through Human-. arXiv preprint arXiv:2603.16567 , year =

    arXiv

  72. [73]

    ACM Conference on Fairness, Accountability, and Transparency , year =

    Moore, Jared and Haber, Nick and others , title =. ACM Conference on Fairness, Accountability, and Transparency , year =

  73. [74]

    Psychiatric Times , year =

    , title =. Psychiatric Times , year =

  74. [75]

    Ethics Sheet for Automatic Emotion Recognition and Sentiment Analysis

    Mohammad, Saif M. Ethics Sheet for Automatic Emotion Recognition and Sentiment Analysis. Computational Linguistics. 2022. doi:10.1162/coli_a_00433

    work page doi:10.1162/coli_a_00433 2022

  75. [76]

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

    Gaps in the safety evaluation of generative AI , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

  76. [77]

    International

    Bengio, Yoshua and Mindermann, S. International. arXiv preprint arXiv:2501.17805 , year=

    arXiv

  77. [78]

    Emotion Concepts and their Function in a Large Language Model , year =

  78. [79]

    and Mehta, R

    Moore, J. and Mehta, R. and Agnew, W. and Anthis, J. R. and Louie, R. and Mai, T. and Yin, D. and Cheng, J. and Paech, A. and Klyman, K. and Chancellor, S. and Lin, J. and Haber, N. and Ong, D. C. , title =. 2025 , howpublished =

    2025

  79. [80]

    European journal of analytic philosophy , volume=

    Autonomy and the emotions , author=. European journal of analytic philosophy , volume=. 2006 , publisher=

    2006

  80. [81]

    2013 , publisher=

    Emotional insight: The epistemic role of emotional experience , author=. 2013 , publisher=

    2013

Showing first 80 references.