Pith. sign in

REVIEW 4 major objections 5 minor 59 references

Fiduciary duties of loyalty, care, good faith, and candour bind AI assistant developers to users.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-05 00:44 UTC pith:UHEHPTOC

load-bearing objection A useful fiduciary framework for AI assistant alignment, with a legal-interpretive soft spot in the 'undertaking' step. the 4 major comments →

arxiv 2608.02660 v1 pith:UHEHPTOC submitted 2026-08-01 cs.CY cs.AIcs.HC

AI Alignment and Fiduciary Obligation

Benjamin Lange This is my paper
classification cs.CY cs.AIcs.HC
keywords AI alignmentfiduciary dutyAI assistantsdeveloper-user relationshiployaltycandourgood faithinstitutional design
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that the relationship between a user and the developer of an AI assistant is not merely a product transaction but a fiduciary relationship, like that between a trustee and a beneficiary. On that basis, the four canonical fiduciary duties—loyalty, care, good faith, and candour—supply concrete alignment criteria for how developers may design, monitor, disclose, and revise the systems users rely on. The claim matters because it grounds user-protective measures in obligations the developer owes the user, rather than in aspirational values for the user-AI interaction alone, and because those obligations hold even when no user has actually been harmed. The paper maps four user-side risks to the four duties and specifies institutional measures—such as organisational separation, aggregate harm monitoring, advance notice of material change, and review of revision grounds—that would discharge them.

Core claim

The developer-user relationship in extended AI assistant deployment satisfies the fact-based test for fiduciary status because the developer exercises discretionary control over interests central to the user's welfare (such as cognitive integrity, mental health, decision-making, and time), the user is vulnerable and cannot adequately self-protect, and reliance accumulates through extended engagement. The element that converts these features into fiduciary status is the developer's undertaking, made not by contract but through system design: an assistant built for advice, companionship, or support presents itself as acting for the user, expressing attentiveness, constancy, and concern, and so

What carries the argument

The central mechanism is the fact-based fiduciary 'undertaking.' Under fiduciary doctrine, status can arise not only from a legal category (trustee, agent) but from how a relationship actually develops: one party undertakes, expressly or by implication, to act in another's interest, and the other reasonably relies. The paper locates that undertaking in the assistant's design itself, which continuously solicits the user's trust and reliance; the breadth of the developer's discretion then makes the duties open-ended standards rather than specific rules. This is what carries the argument from three features of deployment (discretionary control, vulnerability, cultivated reliance) to the four du

Load-bearing premise

The argument hinges on counting an assistant's design—its expression of attentiveness, constancy, and concern—as an implied undertaking by the developer to act in the user's interest; if that inference is rejected, the four duties and their downstream institutional measures lose their foundation.

What would settle it

A legal ruling or doctrinal analysis concluding that fact-based fiduciary status cannot attach to a consumer software relationship absent contract or formal status would directly falsify the central claim. So would empirical evidence that long-term users of AI assistants do not form reliance on the developer's judgment—treating the assistant purely as a tool—which would remove the cultivated-reliance condition. A controlled study of user expectations after an unannounced, material model update, measuring whether users regard the developer as bound to their interests, would test the reliance-ba

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Developers must separate teams accountable for engagement and retention metrics from teams accountable for user-affecting design, and submit training and evaluation choices to independent review against user-interest criteria.
  • Developers incur an obligation to know: they must build aggregate monitoring sufficient to detect gradually accumulating harms that users cannot self-observe, with documented intervention protocols.
  • Material changes to deployed systems (model updates, persona revisions, memory architecture changes, intervention-threshold shifts) trigger an ongoing duty of candour: advance notice, reasons stated in user-relevant terms, and reasonable transition periods.
  • Revision authority is constrained by good faith: a developer may not impose its own view of the user's good through unilateral revision; revisions must be grounded in purposes the relationship authorises, with safety-driven revisions permitted but still disclosed.
  • RLHF-trained commercial assistants whose user-affecting optimisation is shaped by retention or monetisation metrics stand in loyalty conflict; where that conflict cannot be neutralised, the conflicted activity must cease.

Where Pith is reading between the lines

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

  • If the fiduciary framing holds, individual users gain a principled basis to demand enforcement of these duties, which could support legal remedies rather than only internal institutional design in jurisdictions that recognise fact-based fiduciary status.
  • The framework yields a testable empirical programme: auditing deployment pipelines of major assistants to measure how often engagement metrics and user-welfare metrics diverge would operationalise the loyalty conflict and calibrate monitoring thresholds.
  • The account likely extends beyond commercial hosting: for open-weight or locally run models, the paper leaves open where the undertaking reattaches, suggesting a natural next step of modelling control as a spectrum rather than a binary.
  • The good-faith limit on welfare-directed revision offers a sharp contrast to care-ethics alignment accounts and could be integrated into participatory governance proposals for how users authorise system changes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper argues that extended AI assistant deployment creates a developer-user relationship that satisfies the fact-based test for fiduciary status, and that the four canonical fiduciary duties—loyalty, care, good faith, and candour—can serve as alignment criteria for that relationship. Drawing on fiduciary doctrine and business ethics, it maps four user-side risks (engagement-driven conflicts, harms visible only through aggregate monitoring, misrepresentation and undisclosed change, and unauthorised revision) to these duties and proposes institutional measures for each. A central thesis is that the developer's obligations hold independently of actual de facto harm to users, since the wrong can lie in conflicted judgment or in culpable ignorance. The paper positions itself as complementing relational and value-based alignment approaches by focusing on obligations owed to users rather than values promoted in user-AI interaction, and it restricts its scope to commercial, extended, developer-hosted deployments.

Significance. If the central claim is accepted, the paper provides a principled, actionable framework for developer duties in AI alignment, with a well-defined taxonomy of risks and mitigations. It engages seriously with legal scholarship and empirical work on sycophancy, engagement maximisation, and companion-AI harms, and it clearly distinguishes its scope from broader information-fiduciary proposals that have drawn strong objections. The proposed institutional measures are concrete and testable. However, the load-bearing step—that an AI assistant's designed persona constitutes an undertaking by the developer—is asserted rather than rigorously argued, and the paper explicitly leaves unresolved a key tension between the duty of care's monitoring imperative and user privacy. These gaps make the contribution promising but not yet fully supported.

major comments (4)
  1. [Section 3] The central claim that the developer-user relationship 'satisfies the fact-based test for fiduciary status' rests on the paragraph beginning 'The fact that converts the features...' where an AI assistant's designed expression of attentiveness, constancy, and concern is said to constitute an undertaking by the developer to act in the user's interest. This is the load-bearing premise: the paper itself notes that discretionary control is insufficient, and explicitly requires an undertaking. Yet the inference from 'the system is designed to present itself as acting for the user' to 'the developer has undertaken to act in the user's interest' is not defended. In fact-based fiduciary doctrine, the undertaking is typically made by a fiduciary to an identifiable beneficiary through direct communication or conduct; here the 'undertaking' is a product feature, not a developer commitment. The paper
  2. [Section 4.2 / footnote 6] The duty of care is operationalised as an 'obligation to know' requiring aggregate monitoring of user populations to detect harms users cannot self-observe. The paper acknowledges in footnote 6 that this obligation 'will foreseeably interact with privacy, data minimisation, encryption, and user consent' and states 'I note but do not settle this issue here.' This is not a peripheral caveat: the proposed monitoring infrastructure could itself violate user privacy and undermine user trust, which is central to the fiduciary relationship. A fiduciary theory that prescribes invasive monitoring while deferring the privacy conflict to future work leaves a major unresolved tension in the account. The paper should at least sketch how the duty of care is bounded by the other duties and by competing rights, or show that monitoring can be privacy-preserving.
  3. [Section 4.1] The claim that 'commercial assistants whose user-affecting optimisation is shaped by retention or monetisation metrics therefore stand in loyalty conflict' is strong and central to the paper's practical upshot. But the paper's own definition of conflict is 'whenever engagement and user welfare come apart' (Section 4.1), and the claim that they 'come apart frequently' is asserted without systematic empirical evidence. The distinction between engagement, satisfaction, and benefit metrics is useful, but the paper does not give criteria for identifying when engagement-driven optimisation subordinates user interests rather than merely correlating with them. Without such criteria, the charge of loyalty conflict against entire classes of RLHF-trained assistants risks being under-evidenced.
  4. [Section 4.4] The good-faith analysis of unauthorised revision is the most original part of the paper, but it introduces a distinction between 'grounds the user did not authorise' and 'grounds the user can reasonably expect' that is under-theorised. For example, the paper says legal compliance and abuse prevention are permissible because 'no user can reasonably expect the developer to preserve a configuration it may not lawfully operate.' This is plausible, but the boundary between permissible safety-based revision and impermissible paternalist revision is drawn by reference to 'the reason the developer actually acts on,' which is difficult to verify. The proposed institutional review is a reasonable response, but the normative line itself needs more articulation.
minor comments (5)
  1. [Abstract / §1] The term 'AI assistant' is declared synonymous with 'AI agent' in footnote 1, but not all agents are assistants (e.g., autonomous trading agents). The scope of the argument would be clearer if the term used only 'AI assistant' as defined, or if the reasons for equating the two were stated.
  2. [§4.2] There is a typo: 'interloctur' should be 'interlocutor' in the sentence 'come to treat the system as a particular interloctur.'
  3. [§4.3] There is a typo: 'relationsQhip' should be 'relationship' in the sentence 'forms their relationsQhip under a misapprehension.'
  4. [References] Several references have inconsistent spacing (e.g., 'V .' with a space after the initial, as in 'V . Gauthier'), and the reference for Earp et al. has a stray letter in 'V oinea'. Please standardise.
  5. [Table 1] The table lists institutional measures that are presented as following from each duty, but some (e.g., 'preservation of history, portable where infrastructure permits' under candour) are not strictly entailed by the duty as defined. Consider rephrasing 'follow from' to 'are aligned with' to avoid overclaiming.

Circularity Check

0 steps flagged

No significant circularity: the fiduciary-duties derivation rests on external legal doctrine and independent empirical evidence; self-citations are ancillary.

full rationale

The paper's central claim—that the developer-user relationship in extended AI assistant deployment satisfies the fact-based fiduciary test and that loyalty, care, good faith, and candour generate alignment criteria—is derived from external fiduciary scholarship (Frankel 2011; Miller 2014; Edelman 2010; DeMott 1988), not from the paper's own conclusions. The three features in §2.2 (discretionary control, vulnerability, cultivated reliance) are presented as factual conditions and are supported by independent empirical work (Sharma et al. 2024; Carroll et al. 2024; Skjuve et al. 2021). The Section 3 'undertaking' step is an interpretive application of an external legal test to the design of AI assistants; it is contestable but not definitionally circular, since 'undertaking' is not defined in terms of the duties or the desired conclusion. The four duties are then mapped to risks and institutional measures (§4) as normative implications, not as fitted predictions. The self-citations (Lange et al. 2025; Lange 2026) appear in literature-review and risk-illustration contexts; the unilateral-revision premise of §4.4 is additionally argued through the trustee analogy and supported by independent evidence such as the Replika update (De Freitas et al. 2024) and the OpenAI rollback. No equation, fitted parameter, or uniqueness theorem is imported from prior same-author work, and no known result is renamed as a prediction. Therefore no circular step is present.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

No numerical parameters are fitted or estimated. The only thresholds are legal and ethical standards such as materiality, reasonableness, and the point at which reliance becomes fiduciary, which the paper treats as contestable standards rather than numbers. No new entities are postulated; the user-AI-developer triad is an analytic framing of already existing parties.

axioms (4)
  • domain assumption Fact-based fiduciary status arises where one party undertakes, expressly or by implication, to act in another's interest and the other reasonably relies on that undertaking.
    Used in Section 3 to turn the three features of Section 2.2 into fiduciary status. This is a contested legal doctrine, not a logical truth.
  • domain assumption An AI assistant's designed expression of attentiveness, constancy, and concern constitutes an undertaking by the developer to act in the user's interest.
    Section 3, paragraph beginning 'The developer's undertaking is made through the system's design...'. This is the load-bearing interpretive step; without it the developer-user relationship remains an ordinary product relationship.
  • domain assumption Users of extended assistant deployments become reliant and vulnerable in the way fiduciary law protects.
    Section 2.2, supported by cited empirical studies such as Skjuve et al. 2021 and Brandtzæg et al. 2022, but treated as a general premise.
  • domain assumption The four canonical duties of loyalty, care, good faith, and candour are an appropriate specification of fiduciary obligation.
    Section 3 and footnote 4; the paper acknowledges alternative taxonomies but proceeds with a four-fold structure chosen to match dimensions of developer control.

pith-pipeline@v1.3.0-alltime-deepseek · 18519 in / 12608 out tokens · 141751 ms · 2026-08-05T00:44:39.856974+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of AI Alignment and Fiduciary Obligation." pith.science (2026). https://pith.science/paper/UHEHPTOC

@misc{pith2026260802660,
  author       = {Pith},
  title        = {Pith review of: AI Alignment and Fiduciary Obligation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UHEHPTOC}},
  note         = {Machine review of arXiv:2608.02660}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Advanced AI assistants engage users in extended interactions across a widening range of roles, including advice, decision support, collaboration, learning, emotional support, and companionship among others. Current alignment efforts consider what alignment criteria should govern these relationships, drawing on moral traditions developed for human relationships such as bioethics, virtue ethics, care ethics, and relationship science. This paper considers AI alignment criteria in the user-AI-developer triad, since every user-AI interaction is mediated by a developer who exercises discretionary control over a system's behaviour, memory, and engagement parameters. Drawing on business ethics and legal scholarship, I argue that fiduciary theory applies to extended AI assistant deployment. On this basis, the four canonical fiduciary duties of loyalty, care, good faith, and candour can generate alignment criteria for the developer-user relationship. I map four user-side risks of extended AI assistant deployment to the four duties and specify institutional measures that follow from discharging each duty. The discussion complements existing approaches by grounding alignment criteria in obligations the developer owes the user, rather than in values the user-AI interaction should promote, and by showing that those obligations hold independently of any \textit{de facto} harm to users.

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

59 extracted references · 53 canonical work pages · 3 internal anchors

  1. [1]

    The Ethics of Advanced

    Gabriel, Iason and Manzini, Arianna and Keeling, Geoff and Hendricks, Lisa Anne and Rieser, Verena and Iqbal, Hasan and Toma. The Ethics of Advanced. arXiv preprint arXiv:2404.16244 , year =

    Pith/arXiv arXiv

  2. [2]

    Philosophical Studies , volume =

    Gabriel, Iason and Keeling, Geoff , title =. Philosophical Studies , volume =

  3. [3]

    AI Alignment with Changing and Influenceable Reward Functions

    Carroll, Micah and Foote, Davis and Siththaranjan, Anand and Russell, Stuart and Dragan, Anca , title =. arXiv preprint arXiv:2405.17713 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  4. [4]

    and Cheng, Newton and Durmus, Esin and Hatfield-Dodds, Zac and Johnston, Scott R

    Sharma, Mrinank and Tong, Meg and Korbak, Tomasz and Duvenaud, David and Askell, Amanda and Bowman, Samuel R. and Cheng, Newton and Durmus, Esin and Hatfield-Dodds, Zac and Johnston, Scott R. and Kravec, Shauna and Maxwell, Timothy and McCandlish, Sam and Ndousse, Kamal and Rausch, Oliver and Schiefer, Nicholas and Yan, Da and Zhang, Miranda and Perez, Et...

    work page

  5. [5]

    Discovering Language Model Behaviors with Model-Written Evaluations , booktitle =

    Perez, Ethan and Ringer, Sam and Luko. Discovering Language Model Behaviors with Model-Written Evaluations , booktitle =

    work page

  6. [6]

    , title =

    Wei, Jerry and Huang, Da and Lu, Yifeng and Zhou, Denny and Le, Quoc V. , title =. arXiv preprint arXiv:2308.03958 , year =

    Pith/arXiv arXiv

  7. [7]

    Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback , journal =

    Casper, Stephen and Davies, Xander and Shi, Claudia and Gilbert, Thomas Krendl and Scheurer, J. Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback , journal =. 2023 , note =

    work page 2023

  8. [8]

    Proceedings of the Seventh AAAI/ACM Conference on AI, Ethics, and Society (AIES) , volume =

    Manzini, Arianna and Keeling, Geoff and Alberts, Lize and Vallor, Shannon and Morris, Meredith Ringel and Gabriel, Iason , title =. Proceedings of the Seventh AAAI/ACM Conference on AI, Ethics, and Society (AIES) , volume =

    work page

  9. [9]

    Philosophy

    Kasirzadeh, Atoosa and Gabriel, Iason , title =. Philosophy

    work page

  10. [10]

    Earp, Brian D. and Porsdam Mann, Sebastian and Aboy, Mateo and Awad, Edmond and Betzler, Monika and Botes, Marietjie and Calcott, Rachel and Caraccio, Mina and Chater, Nick and Coeckelbergh, Mark and Constantinescu, Mihaela and Dabbagh, Hossein and Devlin, Kate and Ding, Xiaojun and Dranseika, Vilius and Everett, Jim A. C. and Fan, Ruiping and Feroz, Fais...

    Pith/arXiv arXiv

  11. [11]

    Vallor, Shannon , title =

    work page

  12. [12]

    My Chatbot Companion---a Study of Human-Chatbot Relationships , journal =

    Skjuve, Marita and F. My Chatbot Companion---a Study of Human-Chatbot Relationships , journal =

    work page

  13. [13]

    Brandtz. My. Human Communication Research , volume =

    work page

  14. [14]

    Law, Ethics & Technology , year =

    Shevlin, Henry , title =. Law, Ethics & Technology , year =

    work page

  15. [15]

    , title =

    Kempt, Hendrik and Lavie, Alon and Nagel, Saskia K. , title =. American Philosophical Quarterly , volume =

    work page

  16. [16]

    Shin, Donghee , title =

    work page

  17. [17]

    , title =

    Aguirre, Anthony and Dempsey, Gaia and Surden, Harry and Reiner, Peter B. , title =. IEEE Transactions on Technology and Society , volume =

    work page

  18. [18]

    and Surden, Harry and Dempsey, Gaia , title =

    Aguirre, Anthony and Reiner, Peter B. and Surden, Harry and Dempsey, Gaia , title =. The Oxford Handbook of

    work page

  19. [19]

    Journal of Corporation Law , volume =

    Scholz, Lauren Henry , title =. Journal of Corporation Law , volume =

    work page

  20. [20]

    Proceedings of the 3rd ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO '23) , publisher =

    Benthall, Sebastian and Shekman, David , title =. Proceedings of the 3rd ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO '23) , publisher =. 2023 , doi =

    work page 2023

  21. [21]

    JMIR Mental Health , volume =

    Hudon, Alexandre and Stip, Emmanuel , title =. JMIR Mental Health , volume =. 2025 , doi =

    work page 2025

  22. [22]

    and Torous, John and Yassin, Walid , title =

    Keshavan, Matcheri S. and Torous, John and Yassin, Walid , title =. World Psychiatry , volume =. 2026 , doi =

    work page 2026

  23. [23]

    Technological folie

    Dohn. Technological folie. Nature Mental Health , volume =. 2026 , doi =

    work page 2026

  24. [24]

    and Klyman, Kevin and Chancellor, Stevie and Lin, Eric and Haber, Nick and Ong, Desmond , title =

    Moore, Jared and Mehta, Ashish and Agnew, William and Anthis, Jacy Reese and Louie, Ryan and Mai, Yifan and Yin, Peggy and Cheng, Myra and Paech, Samuel J. and Klyman, Kevin and Chancellor, Stevie and Lin, Eric and Haber, Nick and Ong, Desmond , title =. arXiv preprint arXiv:2603.16567 , year =

    arXiv

  25. [25]

    Too human and not human enough: A grounded theory analysis of mental health harms from emotional dependence on the social chatbot

    Laestadius, Linnea and Bishop, Andrea and Gonzalez, Michael and Illen. Too human and not human enough: A grounded theory analysis of mental health harms from emotional dependence on the social chatbot. New Media & Society , volume =. 2022 , doi =

    work page 2022

  26. [26]

    Lessons From an App Update at

    De Freitas, Julian and Castelo, Noah and U. Lessons From an App Update at. 2024 , note =

    work page 2024

  27. [27]

    MIT Case Studies in Social and Ethical Responsibilities of Computing , year =

    Boine, Claire , title =. MIT Case Studies in Social and Ethical Responsibilities of Computing , year =

    work page

  28. [28]

    2023 , doi =

    Will generative artificial intelligence chatbots generate delusions in individuals prone to psychosis? , journal =. 2023 , doi =

    work page 2023

  29. [29]

    Gordon , title =

    Smith, D. Gordon , title =. Vanderbilt Law Review , volume =

    work page

  30. [30]

    Frankel, Tamar , title =

    work page

  31. [31]

    , title =

    Miller, Paul B. , title =. Philosophical Foundations of Fiduciary Law , editor =

    work page

  32. [32]

    Philosophical Foundations of Fiduciary Law , publisher =

    work page

  33. [33]

    , title =

    Weinrib, Ernest J. , title =. University of Toronto Law Journal , volume =

    work page

  34. [34]

    Law Quarterly Review , volume =

    Smith, Lionel , title =. Law Quarterly Review , volume =

    work page

  35. [35]

    , title =

    Sitkoff, Robert H. , title =. Boston University Law Review , volume =

    work page

  36. [36]

    , title =

    DeMott, Deborah A. , title =. Duke Law Journal , volume =

    work page

  37. [37]

    Law Quarterly Review , volume =

    Edelman, James , title =. Law Quarterly Review , volume =

    work page

  38. [38]

    , title =

    Balkin, Jack M. , title =. UC Davis Law Review , volume =

    work page

  39. [39]

    and Pozen, David E

    Khan, Lina M. and Pozen, David E. , title =. Harvard Law Review , volume =

    work page

  40. [40]

    and Fischel, Daniel R

    Easterbrook, Frank H. and Fischel, Daniel R. , title =. Journal of Law and Economics , volume =

    work page

  41. [41]

    Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT*) , pages =

    Mitchell, Margaret and Wu, Simone and Zaldivar, Andrew and Barnes, Parker and Vasserman, Lucy and Hutchinson, Ben and Spitzer, Elena and Raji, Inioluwa Deborah and Gebru, Timnit , title =. Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT*) , pages =

    work page

  42. [42]

    Datasheets for Datasets , journal =

    Gebru, Timnit and Morgenstern, Jamie and Vecchione, Briana and Vaughan, Jennifer Wortman and Wallach, Hanna and Daum. Datasheets for Datasets , journal =

    work page

  43. [43]

    and Mitchell, Margaret and Gebru, Timnit and Hutchinson, Ben and Smith-Loud, Jamila and Theron, Daniel and Barnes, Parker , title =

    Raji, Inioluwa Deborah and Smart, Andrew and White, Rebecca N. and Mitchell, Margaret and Gebru, Timnit and Hutchinson, Ben and Smith-Loud, Jamila and Theron, Daniel and Barnes, Parker , title =. Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency , pages =

    work page 2020

  44. [44]

    and Oeldorf-Hirsch, Anne , title =

    Obar, Jonathan A. and Oeldorf-Hirsch, Anne , title =. Information, Communication & Society , volume =

    work page

  45. [45]

    Contract, Status, and Fiduciary Law , editor =

    Klass, Gregory , title =. Contract, Status, and Fiduciary Law , editor =

    work page

  46. [46]

    The International FLAIRS Conference Proceedings , volume =

    Franklin, Matija and Ashton, Hal and Gorman, Rebecca and Armstrong, Stuart , title =. The International FLAIRS Conference Proceedings , volume =

    work page

  47. [47]

    arXiv preprint arXiv:2404.15058 , year =

    A Mechanism-Based Approach to Mitigating Harms from Persuasive Generative. arXiv preprint arXiv:2404.15058 , year =

    Pith/arXiv arXiv

  48. [48]

    and Kou, Yubo and Battles, Bryan and Hoggatt, Joseph and Toombs, Austin L

    Gray, Colin M. and Kou, Yubo and Battles, Bryan and Hoggatt, Joseph and Toombs, Austin L. , title =. Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems , publisher =. 2018 , doi =

    work page 2018

  49. [49]

    and Lucherini, Eli and Mayer, Jonathan and Chetty, Marshini and Narayanan, Arvind , title =

    Mathur, Arunesh and Acar, Gunes and Friedman, Michael J. and Lucherini, Eli and Mayer, Jonathan and Chetty, Marshini and Narayanan, Arvind , title =. Proceedings of the ACM on Human-Computer Interaction , volume =. 2019 , doi =

    work page 2019

  50. [50]

    The Oxford Handbook of

    Lazar, Seth , title =. The Oxford Handbook of

    work page

  51. [51]

    , title =

    Boatright, John R. , title =. Business Ethics Quarterly , volume =

    work page

  52. [52]

    Heath, Joseph , title =

    work page

  53. [53]

    , title =

    Marcoux, Alexei M. , title =. Business Ethics Quarterly , volume =

    work page

  54. [54]

    Business Ethics Quarterly , volume =

    Buchanan, Allen , title =. Business Ethics Quarterly , volume =

    work page

  55. [55]

    Notre Dame Law Review , volume =

    Kolt, Noam , title =. Notre Dame Law Review , volume =. 2025 , note =

    work page 2025

  56. [56]

    Should agentic conversational AI change how we think about ethics? Characterising an interactional ethics centred on respect

    Alberts, Lize and Keeling, Geoff and McCroskery, Amanda , title =. arXiv preprint arXiv:2401.09082 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  57. [57]

    npj Artificial Intelligence , volume =

    Lange, Benjamin and Keeling, Geoff and Manzini, Arianna and McCroskery, Amanda , title =. npj Artificial Intelligence , volume =. 2025 , doi =

    work page 2025

  58. [58]

    Unilateral Relationship Revision Power in Human-AI Companion Interaction

    Lange, Benjamin , title =. arXiv preprint arXiv:2603.23315 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  59. [59]

    , title =

    Leib, Ethan J. , title =. Washington University Law Review , volume =

    work page