{"id":"53e9c40f-d382-49d5-be64-820661ae3f52","arxiv_id":"2507.10786","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Through 19 interviews, U.S. smart-home and chatbot users showed context-dependent privacy and security concerns about domestic social robots, expecting transparent indicators, granular controls, and context-appropriate functionality.","lead":"What privacy and security worries do U.S. users of smart homes and chatbots have about domestic social robots? Interviews with 19 people found that concerns shift by context: children trigger the most caution, medical use raises reliability worries, and users want visible controls and data-use transparency.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'highly context-dependent' finding is confounded: all participants saw the seven scenarios in the same fixed order, with purpose scenarios explicitly conditioned on earlier recipient scenarios, so the cross-context comparisons in §4.2 may reflect order and carryover effects rather than genuine…","rationale":"The reader's weakest assumption was sample representativeness. That concern is real but well-managed: the paper scopes most claims to U.S.-based participants with smart-home and chatbot experience and lists demographic limitations in §3.1. The more load-bearing threat is internal validity of the paper's comparative, context-dependent contribution. For the central claim to hold, the between-scenario differences in §4.2 must reflect the manipulated context. Because scenario order was fixed and purpose scenarios explicitly referenced the preceding recipient scenarios, the data cannot distinguish context effects from order and carryover effects. The paper deserves credit for transparent reporting, two-coder coding with disagreement resolution, a saturation check, IRB-approved deception with debriefing, and public supplementary materials. Those strengths support the descriptive inventory of concerns, but they do not establish that the concerns are 'highly context-dependent' in the comparative sense the discussion asserts. I therefore recommend conditional acceptance: keep the descriptive findings, but either provide an order analysis of the existing transcripts or a counterbalanced replication before claiming cross-context differences; otherwise, frame the comparative patterns as exploratory and adjust the abstract and discussion accordingly.","tokens_in":25692,"tokens_out":6277,"duration_ms":86872,"concrete_test":"Re-code the existing 19 transcripts at scenario-block level, recording for each code in the final codebook the block in which that code first appeared. Then compute, for the education, medical, and therapy blocks, the fraction of their reported themes that already appeared in the earlier recipient blocks. If that fraction is high—for example, if misinformation and reliability themes are first mentioned during the child, elderly, or household blocks—then the purpose-specific attributions in §4.2 are not separable from scenario ordering and the 'context-dependent' claim is unsupported as stated.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim in §5—'These concerns were highly context-dependent'—requires that the observed differences across device-recipient and purpose scenarios are attributable to the scenario context. The protocol in §3 does not establish that attribution. All 19 participants were walked through the same fixed sequence: self, child, elderly, household, then education, medical, therapy (Appendix A.3.2). Moreover, each purpose scenario instructs participants to 'keep in mind the four user scenarios we just described,' making the purpose ratings conditional on the earlier recipient scenarios rather than independent measurements of purpose. Consequently, the findings that misinformation dominated education, reliability dominated medical, and therapy 'did not introduce new privacy or security concerns' (§4.2) could reflect cumulative exposure, fatigue, or anchoring to earlier blocks instead of real context differences. The limitation paragraph in §3.1 acknowledges that randomizing order 'could' improve the study, but the acknowledgment does not remove the confound from the paper's central comparative contribution. This is an internal-validity problem, not a generalizability problem: even for this sample, the cross-context comparisons are not cleanly identified.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"Based on 19 semi-structured interviews with U.S.-based Prolific participants who own smart-home devices and have used AI chatbots, the paper examines privacy and security perceptions of a hypothetical domestic social robot. Participants were shown a six-feature robot specification and seven scenarios (four device-recipient contexts and three purpose contexts), then asked about their concerns and expectations. The authors report that participants largely anchored social robots to smart speakers and chatbots, worried most about audio/video data collection and data inference, showed context-dependent concern patterns (e.g., child safety, elderly usability, shared-household data leakage, misinformation in education, reliability in medical use), and expected transparency, tangible controls, parental controls, and regulation. The paper contributes qualitatively derived design and policy implications for domestic social robots in the U.S.","tokens_in":25815,"tokens_out":5873,"duration_ms":71247,"significance":"If the findings are accepted, this is a useful early qualitative map of consumer privacy and security attitudes toward domestic social robots, with concrete design guidance such as on-device indicators, kill switches, parental controls, and regulatory clarification. The study has notable strengths: the protocol was IRB-approved, the deception was disclosed and debriefed, saturation is described with four additional interviews, two researchers coded independently, and the reporting uses a defined frequency terminology. The materials are openly linked. The main risk is the internal validity of the cross-context comparisons, which underpin the paper's central claim that concerns were 'highly context-dependent.'","major_comments":[{"comment":"The central comparative claim in §5 that concerns were 'highly context-dependent' is not cleanly identified. All 19 participants experienced the seven scenarios in the same fixed order (self, child, elderly, household, then education, medical, therapy), and the three purpose scenarios explicitly instructed participants to 'keep in mind the four user scenarios we just described' (Appendix A.3.2). Consequently, the cross-context differences reported in §4.2 (e.g., misinformation dominating education, reliability dominating medical, therapy introducing no new concerns) could reflect order effects, fatigue, or anchoring to earlier blocks rather than the scenario content. The limitation paragraph in §3.1 acknowledges that randomizing order 'could' improve the study, but the results are still presented as comparative findings. Please either reframe these comparisons as exploratory and hypothesis-generating, provide transcript-based evidence (e.g., order-of-mention analyses or participants' explicit cross-scenario comparisons) that order/carryover did not drive the pattern, or supplement with data from a randomized-order protocol.","section":"§3, Appendix A.3.2, §4.2, §5"},{"comment":"The abstract and conclusion generalize to 'U.S. users' security and privacy needs and concerns, but the sample consists of 19 Prolific participants who all own smart-home devices, all have used chatbots, are mostly highly educated, and are concentrated in the 25-64 age range (Table 1). The limitations section acknowledges this demographic narrowness, but the headline claim is not correspondingly qualified. Please align the abstract and conclusion with the RQs' more precise framing ('U.S.-based participants') and explicitly state that the design implications are grounded in this sample of smart-home and chatbot users, pending broader validation.","section":"Abstract, §5, §3.1"},{"comment":"The statement in §4.2 that 'no participants talked about unique risks introduced by the AI components of social robots' is used in §5 to suggest 'a potential lack of awareness' of generative-AI risks. This absence claim is not well supported by the protocol: the interview did not explicitly prompt for AI-specific risks until the final section (§A.3.3), and earlier sections asked open-endedly about comfort and concerns. The absence of unprompted mentions may reflect the interview's sequencing and the participants' limited familiarity with social robots rather than a genuine lack of awareness. Please soften this claim or provide evidence about what participants were asked before concluding that they lacked awareness of AI-specific risks.","section":"§4.2, §5"}],"minor_comments":[{"comment":"In 'Section 1: Knowledge and Awareness Toward Social Robots', the text says 'In the second section, we asked questions...' but this is the first section; the numbering appears to be a typo.","section":"§3"},{"comment":"In the paragraph beginning 'Our study builds upon existing research,' 'laregely' should be 'largely.'","section":"§2"},{"comment":"In the interview scenario design, 'This scenarios focuses on a child' should read 'This scenario focuses on a child.'","section":"§3"},{"comment":"The name 'Shcafer et al.' in the discussion of regulatory work should be 'Schafer et al.,' matching reference [101].","section":"§5"},{"comment":"The limitations paragraph discusses the fixed scenario order as if it were solely a flow preference, but the conditioning of purpose scenarios on recipient scenarios is also a design choice that affects interpretation; the acknowledgment could be more specific about the carryover mechanism.","section":"§3.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for a CS/CY venue and offers a useful qualitative dataset. The main risk is the order/carryover confound in the cross-context comparisons that support the central claim; I would support publication after the authors reframe the comparative findings as exploratory or provide additional evidence. I do not see citation or novelty problems, and the open materials are a strength."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version: this is a worthwhile qualitative study that fills a real gap, but its central comparative claim is shakier than the authors suggest. The fixed scenario order and the conditional wording of the purpose questions mean the 'highly context-dependent' finding in §5 is not cleanly identified.\n\nWhat's new and what it does well: the multi-context design—four recipient groups by three purposes—is a legitimate extension beyond the single-context studies that dominate this literature. The observation that participants overlooked AI-specific risks like memorization, and expected HIPAA-level protection for medical use, is genuinely useful for designers and regulators. The method is careful: IRB-approved deception with debriefing, saturation described, two independent coders, and GitHub materials that make the protocol and codebook publicly available. That is real evidence and deserves credit.\n\nThe soft spot: the stress-test note is right. All 19 participants saw the same sequence—self, child, elderly, household, then education, medical, therapy—and the purpose scenarios explicitly told them to keep the earlier recipient scenarios in mind. So the cross-context comparisons in §4.2, such as misinformation dominating education and reliability dominating medical, could just as easily reflect carryover, fatigue, or anchoring as genuine context differences. The limitation paragraph admits that randomizing 'could' improve the study, but the acknowledgment does not rescue the central comparative claim. The other limitations—no inter-rater reliability, a small and highly educated Prolific sample, and an abstract that says 'U.S. users'—are real but secondary.\n\nWho this is for: HRI and privacy researchers, plus designers and regulators working on consumer robots. The descriptive themes and design recommendations stand even if the comparative framing is weakened. The paper deserves a serious referee; the referee should push for a randomized follow-up or for language in the discussion that treats the comparative findings as exploratory rather than established.","headline":"A useful qualitative map of privacy expectations for domestic social robots, but the central 'highly context-dependent' claim is undermined by a fixed scenario order the authors acknowledge but do not remove.","tokens_in":26410,"tokens_out":2005,"would_cite":true,"duration_ms":25279,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Privacy fears for home robots hinge on who uses them, interviews show","keywords":["social robots","privacy concerns","security concerns","context-dependent privacy","smart home","qualitative interviews","domestic robots","data transparency"],"falsifier":"A representative survey of several hundred U.S. adults, asked the same recipient-by-purpose scenario questions, would settle whether concerns are as context-dependent and expectations as uniform as the 19 interviews suggest.","tokens_in":25432,"feed_emoji":"🤖","tokens_out":3773,"duration_ms":45343,"temperature":0.7,"pith_summary":"This paper argues that U.S. consumers who already use smart home devices and chatbots hold context-dependent privacy and security expectations for domestic social robots. Through 19 semi-structured interviews, the authors find that concerns shift sharply with the intended user (self, child, elderly, or household) and the purpose (education, medical, therapy). They also show that participants expect tangible privacy controls, clear data-collection indicators, and context-appropriate functionality, and that current U.S. regulations such as HIPAA and COPPA do not yet provide these protections. The study matters because social robots are still early in commercialization, leaving a design and policy window that this work tries to inform.","feed_headline":"Privacy fears for home robots hinge on who uses them","feed_subtitle":"19 U.S. smart-home users show concerns shift by user and purpose, and current privacy law lags behind.","key_machinery":"The central instrument is a scenario-based semi-structured interview built on a six-feature specification of a domestic social robot: visual recognition, voice recognition, expressive communication, personalization, navigation and mapping, and internet connection. This specification was compiled as a superset of capabilities from five commercial robots, and participants were told it was a real product to ground their reactions. Four device-recipient conditions (self, child, elderly, household) and three purpose-of-use conditions (education, medical, psychological therapy) were systematically walked through, and two researchers independently coded the responses using thematic analysis. This recipient-by-purpose matrix is what lets the paper attribute differences in concern and expectations to context rather than to the device itself.","core_discovery":"The paper claims that consumers' privacy and security concerns about domestic social robots are not uniform but highly context-dependent. Participants were least worried about owning a robot for themselves, most worried about children's data and social development, concerned about elderly users' usability and susceptibility to manipulation, and focused on misinformation in educational uses and reliability in medical uses. Nearly all participants voiced concerns about audio and video data collection and data inference before being prompted, yet few raised AI-specific risks such as model memorization. The authors further claim that participants expect transparency through multiple channels, on-device signals of data collection, review-and-delete functions, physical kill switches, and parental controls, and that these expectations are not met by current devices or the U.S. regulatory landscape, which treats social robots largely like ordinary IoT devices.","pith_inferences":["If context-dependence holds at scale, a single privacy dashboard or one-time consent flow will not suffice; robots would need mode- or user-specific data policies that adapt to who is present and what the robot is doing.","The privacy resignation expressed by some participants suggests that transparency and controls, while necessary, may not change adoption behavior much unless they also address users' sense of inevitability.","The recipient-by-purpose scenario matrix could be repurposed as a practical risk-assessment checklist for regulators evaluating future social robot products.","A longitudinal field study with actual robots could test whether the stated expectations from interviews match observed privacy-seeking behavior, addressing the privacy-paradox question the paper acknowledges."],"forward_implications":["Designers should build tangible privacy controls, including on-device camera/audio indicators, physical kill switches, review-and-delete functions, and granular parental controls.","Medical-adjacent uses will trigger expectations of HIPAA-like protections that current law does not provide unless the robot is supplied by a covered entity.","If social robots consolidate the smart home into one device, data-linkage risks increase, but centralized privacy management could reduce privacy fatigue.","The low salience of AI-specific risks among participants implies that transparency about model training, inference, and data use is a necessary design input.","U.S. regulations should treat social robots as a category distinct from general IoT, closing the COPPA loophole for data collected about children rather than directly from them."],"supporting_citations":[{"why":"Provides prior survey evidence on adult perceptions of privacy and attitudes toward social robots in the home, the closest baseline for this study.","marker":"[72]"},{"why":"Shows through family co-learning workshops that context and use cases shape privacy decisions, the prior work this interview study extends.","marker":"[73]"},{"why":"Establishes that IoT privacy expectations are heavily influenced by use context, the framework the authors build upon for social robots.","marker":"[85]"},{"why":"Supplies the elderly-user adoption and usability concerns that inform the elderly recipient scenario.","marker":"[92]"},{"why":"Documents parents' child-safety concerns in smart homes, grounding the child recipient scenario.","marker":"[110]"},{"why":"Anchors expectations for smart speaker privacy controls, including review and deletion of voice data, which participants transferred to social robots.","marker":"[69]"},{"why":"Provides the framework of physical, psychological, and social privacy dimensions used to interpret participants' non-data privacy concerns.","marker":"[18]"},{"why":"Offers the foundational analysis of robots' privacy risks under U.S. regulation, which the paper updates for the social robot context.","marker":"[19]"}],"fun_headline_variants":["Privacy fears for home robots shift by user and purpose","Kids' data biggest worry for domestic social robots","Social robots raise privacy concerns current law can't address","Home robot users fear data inference, demand kill switches","Context shapes privacy worries for home robots"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The findings rest on 19 interviews with a mostly highly educated, tech-experienced Prolific sample, and if that group's views do not reflect the wider U.S. public, the design and policy recommendations may not generalize.","fun_headline_variants_meta":{"raw":{"variants":["Privacy fears for home robots shift by user and purpose","Kids' data biggest worry for domestic social robots","Social robots raise privacy concerns current law can't address","Home robot users fear data inference, demand kill switches","Context shapes privacy worries for home robots"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000289,"raw_usage":{"total_tokens":1670,"prompt_tokens":896,"completion_tokens":774,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":512,"completion_tokens_details":{"reasoning_tokens":702}},"tokens_in":512,"tokens_out":774,"duration_ms":9062,"temperature":1.0,"reasoning_tokens":702,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T17:25:48.229504+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A representative survey of several hundred U.S. adults, asked the same recipient-by-purpose scenario questions, would settle whether concerns are as context-dependent and expectations as uniform as the 19 interviews suggest.","supporting_citations":[{"cited_title":"Surveying adult perceptions of pri- vacy and attitudes towards social robots in the home","cited_arxiv_id":null,"evidence_quote":"Provides prior survey evidence on adult perceptions of privacy and attitudes toward social robots in the home, the closest baseline for this study."},{"cited_title":"Our busi- ness, not the robot’s: family conversations about pri- vacy with social robots in the home","cited_arxiv_id":null,"evidence_quote":"Shows through family co-learning workshops that context and use cases shape privacy decisions, the prior work this interview study extends."},{"cited_title":"Privacy expectations and preferences in an IoT world","cited_arxiv_id":null,"evidence_quote":"Establishes that IoT privacy expectations are heavily influenced by use context, the framework the authors build upon for social robots."},{"cited_title":"Analyzing the elderly users’ adoption of smart-home services","cited_arxiv_id":null,"evidence_quote":"Supplies the elderly-user adoption and usability concerns that inform the elderly recipient scenario."},{"cited_title":"Child safety in the smart home: parents’ perceptions, needs, and mitiga- tion strategies","cited_arxiv_id":null,"evidence_quote":"Documents parents' child-safety concerns in smart homes, grounding the child recipient scenario."},{"cited_title":"Alexa, are you listening? privacy percep- tions, concerns and privacy-seeking behaviors with smart speakers","cited_arxiv_id":null,"evidence_quote":"Anchors expectations for smart speaker privacy controls, including review and deletion of voice data, which participants transferred to social robots."}],"review_version":1}