Source-linked AI summary
Who Bears the Cost of Honesty? A FAccT Workshop Synthesis and Research Agenda for Equitable AI Disclosure
Runlong Ye, Jessica He, Finola Finn, Angel Hsing-Chi Hwang, Donal Khosrowi, Seyun Kim, Morgan Klaus Scheuerman
TL;DR
AI disclosure can support accountability, trust, provenance, and consent while also exposing users to suspicion, stigma, and surveillance. This paper synthesizes a workshop using scenario-anchored power mapping and design fiction to examine these tensions across represented domains. Its artifacts depict disclosure as multi-actor accountability, surface risks for accessibility-related AI use, and motivate the Cost-of-Honesty Stack as a diagnostic framework for context-specific governance.
Problem
AI disclosure can provide transparency-related benefits while imposing uneven privacy, stigma, surveillance, and reputational burdens, but these distributional tensions remain context-dependent.
Method
The paper documents and synthesizes a FAccT workshop combining scenario-based power mapping, design fiction, and artifact-centered thematic analysis.
Results
Workshop artifacts represented disclosure as a multi-actor accountability process, raised concerns that accessibility-related AI use could affect workplace evaluations, and explored contextualized disclosure alternatives.
Takeaways & Limitations
The cost-of-honesty lens and Stack provide starting points for future empirical work, participatory policy development, and domain-specific design without presuming one universal disclosure standard.
Takeaways & Limitations
The corpus is small, self-selected, speculative, and nonrepresentative, so its artifacts surface tensions and possible futures rather than establish prevalence.
Abstract
from arXiv · showhide
AI disclosure is increasingly promoted and sometimes required as a route to transparency, accountability, provenance, and trust. Yet disclosure can also expose AI users to suspicion, stigma (e.g., competence penalties), and surveillance, affecting minoritized groups in particular. This paper reports on Who Bears the Cost of Honesty?, a CRAFT workshop at the 2026 ACM Conference on Fairness, Accountability, and Transparency that used scenario-anchored power mapping and design fiction to explore the benefits, harms, tensions, and power asymmetries that emerge under AI disclosure norms and mandates. We document the workshop design and analyze the disclosure approaches participants co-created, comprising four completed power maps, three context cards, and one interface prototype. These artifacts span education, workplace, politics/journalism, and interpersonal contexts. They depict disclosure as a multi-actor accountability process, surface concerns that the use of accessibility-related AI could be held against workers in performance evaluations, and explore how context-specific, bottom-up disclosures may support transparency while mitigating some risks of stigma and misinterpretation. We contribute (1) a documented two-stage workshop method; (2) an artifact-grounded thematic synthesis; and (3) a diagnostic framework, the Cost-of-Honesty Stack, with provisional design suggestions and research directions.
1 Introduction
AI disclosure may support provenance, accountability, and calibrated trust, but it can also impose privacy, stigma, surveillance, and reputational burdens unevenly. The workshop responds by treating disclosure as a situated governance problem shaped by unequal power.
- Disclosure can establish provenance, reduce deception, support accountability, and calibrate trust, but may also expose users to privacy, credit, autonomy, and reputational losses.Workplace competence and motivation penalties, including disproportionate effects on some minoritized users, illustrate these burdens.
- Disclosure requirements are justified by benefits to recipients, institutions, and publics, while the burdens of disclosure often fall on the person revealing AI use.The paper names this distributional question the cost of honesty.
- Consequences and expectations vary across education, workplace, freelance, and other domains, making disclosure a family of situated governance problems rather than a universal labeling question.Evidence includes detector concerns for non-native English writers and contested disclosure requirements for different forms of assistance.
- The workshop used scenario-based power mapping followed by near-future design fiction to explore stakeholders, asymmetries, alternative arrangements, and unresolved harms.Its scenarios covered six domains, while the archived participant corpus represented four.
- The paper contributes a documented workshop method, an artifact-grounded synthesis, and the Cost-of-Honesty Stack as a diagnostic framework with provisional design prompts and research directions.
2 Workshop Design
The workshop combined power mapping and design fiction to move participants from identifying disclosure-related power asymmetries toward co-designing alternative arrangements. Its scenarios and activities emphasized distributed stakeholders, accountability, and practical tensions around disclosure.
- The program comprised an introduction, keynote, 45-minute power-mapping activity, 45-minute design-fiction studio, and group share-out.The workshop sought to identify situated harms and asymmetries, design alternatives, and connect researchers with practitioners.
- Six scenario options made power asymmetries concrete, including education’s authorship tension and workplace transparency without exposing disability-related needs.The archived outcomes did not represent every scenario option.
- The keynote emphasized that wording, context, and process transparency matter, favoring specific accounts of AI’s role and human accountability over blunt labels.It also characterized disclosure burdens as falling disproportionately on less powerful actors.
- Groups mapped primary, secondary, and tertiary actors and annotated relationships to ask who could impose, interpret, benefit from, or bear the consequences of disclosure.The activity intentionally expanded analysis beyond the AI user–recipient dyad.
- Groups selected artifact types such as interfaces, policies, refusal processes, labels, audits, norms, or training materials to design near-future disclosure interventions.
3 Data Corpus and Analysis
The paper analyzes a small workshop corpus through artifact-centered thematic synthesis rather than transcript-based qualitative analysis. The analysis records how participants represented actors, interventions, revealed information, agency, mitigated costs, and residual harms.
- The corpus included 22 participants and outputs from four groups representing education, workplace, politics/journalism, and interpersonal relationships.Creative work and healthcare were excluded because their outputs were not represented in the archived corpus.
- The collected materials comprised four completed power maps, three context cards, and one dating-app interface prototype with three screens and a recipient-facing label.No verbatim transcript or systematic facilitator-note set was collected.
- The synthesis used thematic analysis to develop and refine patterns across qualitative materials.
- Researchers recorded stakeholder positions, relationships, decision-making power, disclosure obligations, proposed interventions, information flows, agency, mitigated costs, residual harms, and absent remedies.They compared recurrence and domain-specific differences across artifacts.
- Prompt materials were retained as elicitation metadata but not coded as participant evidence, while independent review and reconciliation refined themes and supporting evidence.
4 Workshop Outcomes
Workshop artifacts represent disclosure as a multi-actor chain whose consequences extend into evaluation, institutional decisions, infrastructure, and governance. They also explore contextualized disclosure while retaining concerns about stigma, surveillance, and contestability.
- 4.1 Distributed impact across multiple actors: All four power maps placed the AI user among the most directly affected actors and represented disclosure as a chain involving definers, interpreters, decision-makers, and downstream-affected people.Mapped relationships included performance review, hiring, firing, editorial decisions, and institutional participation.
- 4.2 Assistive use as evaluative evidence: Education and interpersonal artifacts linked disclosure with stigma, blame, cheating, shortcuts, shame, and unfair evaluation.
- 4.2 Assistive use as evaluative evidence: Workplace artifacts raised privacy concerns about publicizing disability and uncertainty over distinguishing equalizing AI assistance from other uses.They connected disclosure to possible judgments about competence, effort, authenticity, or character.
- 4.3 Contextual disclosure retained interpretive and governance risks: The interpersonal prototype paired a disclosure toggle and optional explanation with a structured recipient-facing label specifying AI role, content status, initiation, review, and tool version.
- 4.3 Contextual disclosure retained interpretive and governance risks: The Cost-of-Honesty Stack locates where disclosure is demanded, interpreted, recorded, and enforced across five sociotechnical layers.Its diagnostic questions connect actors and governance choices across the system.
- 4.3 Contextual disclosure retained interpretive and governance risks: The prototype proposed the less shaming phrase “AI-assisted” and contextual explanation, but its design intentions were not evaluated effects.
- 4.3 Contextual disclosure retained interpretive and governance risks: Artifacts directed attention to who may collect, retain, interpret, contest, or act on disclosures, while explicit appeal, anti-retaliation, deletion, and redress mechanisms were largely absent.
5 The Cost-of-Honesty Stack
The Cost-of-Honesty Stack diagnoses AI disclosure across five sociotechnical layers, from user labor to governance. Applying it to workplace and interpersonal artifacts surfaces how evaluation, policy, infrastructure, and public governance can shape disclosure risks.
- Five diagnostic layers: The Stack locates where disclosure is demanded, interpreted, recorded, and enforced across five layers of a sociotechnical system.Its layers are user labor, evaluation, institution/platform, infrastructure, and governance/public.
- Five diagnostic layers: At the user labor layer, disclosure concerns the purpose and character of AI use and whether it exposes disability, language background, workload, or identity.The framework distinguishes assistive, generative, accessibility-related, procedural, and deceptive uses.
- Five diagnostic layers: At the evaluation, institution/platform, and infrastructure layers, the Stack asks how AI use is judged, governed, captured, accessed, and protected against biased penalties or surveillance.Questions include evaluator assumptions, policy scope and appeals, data retention, log access, and false positives or negatives.
- Five diagnostic layers: The governance/public layer examines which harms disclosure should prevent, who participates in rule-making, and whether implementation creates legal certainty without social penalties.It also asks how disparate impacts are measured after implementation.
- Tracing participant outputs: In the workplace artifact, the Stack connects disability privacy concerns to performance review, hiring or firing, policy scope, data retention, and unresolved exploitation risks.The artifact does not specify provider data retention or concrete societal and regulatory actions, so the Stack poses questions rather than attributing interpretations to participants.
- Tracing participant outputs: In the interpersonal artifact, “AI-assisted” and a structured label provide contextual provenance, but a judgmental recipient could still penalize the sender.The Stack additionally prompts questions about whether labels are self-reported or record-backed and who sets disclosure rules.
6 Applying the Stack: Provisional Design Suggestions
The paper translates recurring Cost-of-Honesty Stack questions into provisional design suggestions grounded in data justice and design justice. It presents these suggestions as starting points for adaptation and evaluation, not validated standards.
- Design suggestions: Table 2 converts recurring Stack questions into provisional design suggestions focused on how people are made visible, treated, and affected by design decisions.The suggestions align with data justice, design justice, and critiques of transparency-only approaches to accountability and fairness.
- Scope: Because the workshop was small, the suggestions require further adaptation and evaluation rather than serving as validated standards.
7 Research Directions
The paper proposes three research directions: measuring disclosure harms and benefits, designing domain-specific disclosure patterns, and building participatory disclosure governance. It frames the workshop methods as one possible way to surface affected stakeholders and residual harms, while noting that their policy value requires further evaluation.
- Measuring disclosure harms and benefits: Future studies could measure how disclosure granularity, timing, audience, and context relate to trust, competence judgments, authenticity judgments, privacy concerns, and willingness to use AI.Sampling could prioritize groups likely to face disproportionate burdens, including disabled users, non-native speakers, women, older or junior workers, and precarious workers.
- Designing domain-specific disclosure patterns: Future work could develop domain-specific disclosure patterns instead of assuming one universal label.Candidate patterns include assistive-use, editorial-provenance, patient-information, student-learning, and creative-attribution disclosures, plus refusal and reciprocal-disclosure processes.
- Building participatory disclosure governance: Participatory policy development could use power mapping, design fiction, and context cards to surface beneficiaries, at-risk groups, and harms created by proposed disclosure artifacts.
- Building participatory disclosure governance: Whether these workshop methods improve policy design requires evaluation beyond the present workshop.
8 Discussion
The discussion reframes AI disclosure as a question of how burdens and protections are allocated, not merely whether information is present. Participant artifacts and the synthesis show why governance, evaluator practice, retention, appeals, and reciprocal transparency matter alongside interface design.
- Allocating burdens and protections: The cost-of-honesty lens asks how disclosure burdens and protections are allocated when institutions pursue integrity, productivity evaluation, assurance, trust, credibility, or accountability.A one-way demand can make vulnerable users carry the moral and evidentiary burden of a broader sociotechnical transition.
- Cross-domain implications: Participant outputs connect disclosure with cheating, shortcuts, stigma, blame, accessibility-related workplace privacy, distributed journalism accountability, and contextualized interpersonal provenance.The examples span education, workplace, politics/journalism, and interpersonal relationships.
- Governance alongside interface design: The synthesis places governance alongside interface design because policy scope, evaluator practice, data retention, appeal mechanisms, and reciprocal transparency determine how disclosures may be used or contested.The Stack therefore includes purpose, proportionality, refusal, contextualization, and remedy as questions for future design and evaluation; these are propositions, not workshop-established effects.
9 Limitations and Ethics
The paper treats its workshop corpus and design suggestions as exploratory rather than representative or validated. It also identifies privacy, positionality, and participant-involvement considerations for future iterations.
- The corpus is small, self-selected, speculative, and time-constrained, so it surfaces tensions and possible futures rather than proving prevalence.
- Raw photographs and quotations from handwritten artifacts can expose identifying details, motivating redrawn composites, paraphrased examples, short anonymized quotations, and consent-based attribution.
- Table 2's design suggestions are starting points for adaptation and evaluation, not validated standards or evaluated policy recommendations.
- The organizers' responsible-AI, HCI, design, and ethics expertise supports synthesis but may shape which harms are most visible.
- Future work should involve affected users more directly, including students, disabled workers, patients, freelancers, non-native speakers, journalists, educators, and others subject to disclosure policies.
10 Conclusion
The conclusion presents AI disclosure as both a governance mechanism and a source of uneven burdens. The workshop synthesis offers a diagnostic framework and provisional directions without assuming one universal disclosure standard.
- AI disclosure can support accountability, trust, provenance, and consent while creating risks of suspicion, stigma, and surveillance.
- The paper uses the cost of honesty to examine how workshop participants represented disclosure's benefits and burdens.
- The documented method, artifact-grounded themes, Cost-of-Honesty Stack, design suggestions, and research directions are starting points for future empirical work and participatory policy development.
- The paper does not presume a single universal disclosure standard, instead supporting domain-specific design.
A Workshop Activity Materials
The workshop activity materials guide groups from a concrete disclosure scenario to a near-future artifact and context card. They require participants to identify beneficiaries, mitigated and unmitigated costs, and remaining harms.
- The appendix reproduces organizer-authored scenario, artifact-type, instruction, and context-card materials as workshop inputs rather than participant-generated findings.
- Groups choose one scenario card and one artifact-type card, then design a 2029 artifact and complete a context card explaining its benefits and risks.
- The scenario cards provide concrete disclosure tensions and stakeholders, while the artifact cards offer forms such as interfaces, policies, labels, reports, rituals, and training materials.
- The context-card prompts ask who gains trust, clarity, credit, protection, control, time, money, or legitimacy.
- Participants assess which costs of honesty are mitigated, which worsen, who still carries risk, and whether costs shift toward someone with less power.
- Groups also consider additional harms, exclusions, misunderstandings, enforcement problems, and what the artifact still needs.