The DOI matches its Crossref record; 49 of the 49 references it cites were found in a registry or an index; none of the 39 sources its claims cite is checked for support yet.
“It's a Fair Game”, or Is It? Examining How Users Navigate Disclosure Risks and Benefits When Using LLM-Based Conversational Agents
Open accessAbstract
The widespread use of Large Language Model (LLM)-based conversational agents (CAs), especially in high-stakes domains, raises many privacy concerns. Building ethical LLM-based CAs that respect user privacy requires an in-depth understanding of the privacy risks that concern users the most. However, existing research, primarily model-centered, does not provide insight into users’ perspectives. To bridge this gap, we analyzed sensitive disclosures in real-world ChatGPT conversations and conducted semi-structured interviews with 19 LLM-based CA users. We found that users are constantly faced with trade-offs between privacy, utility, and convenience when using LLM-based CAs. However, users’ erroneous mental models and the dark patterns in system design limited their awareness and comprehension of the privacy risks. Additionally, the human-like interactions encouraged more sensitive disclosures, which complicated users’ ability to navigate the trade-offs. We discuss practical design guidelines and the needs for paradigm shifts to protect the privacy of LLM-based CA users.
Overview
CoChat Summary
TL;DR
This paper examines how users of LLM-based conversational agents navigate privacy risks and benefits, combining analysis of real-world ChatGPT conversations with interviews. It finds that users trade off privacy, utility, and convenience, but flawed mental models and dark patterns limit their ability to protect themselves.
Key findings
- In the ShareGPT52K dataset, the authors analyzed 200 conversation threads (10,380 messages) and found 106 conversations that actually contained PII; users disclosed both their own and other people’s information, raising interdependent privacy issues.
- Interviews with 19 LLM-based CA users (18 ChatGPT, 1 Bing chat) showed that disclosure intentions were shaped by perceived CA capability, convenience, perceived data sensitivity, resignation, and perceived risks such as institutional misuse, idea theft, and others finding out.
- Users employed ad-hoc protections—censoring/falsifying data, sanitizing copied inputs, or seeking only general advice—with nearly all participants except P5, P8, P11, and P15 using at least one such measure.
- Most participants lacked awareness of the opt-out control for model training (all but P11, P14, P16, P17, P18), and ChatGPT’s opt-out design contained dark patterns: training opt-out was bundled with disabling chat history, while a harder-to-find form to keep history was found by none of the participants.
Methods
- Study 1: qualitative analysis of the ShareGPT52K dataset (50,496 ChatGPT chat histories); Microsoft Presidio detected PII, narrowing to 30K conversations, from which 200 threads were sampled and coded using contextual integrity and affinity diagramming.
- Study 2: semi-structured interviews with 19 LLM-based CA users (18 ChatGPT, 1 Bing chat), recruited via Prolific and social networks, conducted remotely July–August 2023, lasting about 60–90 minutes with $30 compensation; included mental-model drawing and bottom-up coding yielding 62 codes in 6 themes.
- Analysis combined dataset coding and interview coding to address real-world disclosure behaviors, reasons for disclosure or avoidance, and awareness of privacy controls.
Limitations
- The paper states dataset limitations: conversation context was sometimes unclear, users’ thought processes were not directly observable, and ShareGPT extension users may not represent all LLM users.
- The paper states interview limitations: participants may have avoided discussing sensitive conversations, more sensitive sharers may not have participated, and recall or social desirability bias may affect reports.
- The paper also notes broader limitations: focus on ChatGPT, a general-population sample, qualitative rather than quantitative modeling, and the need for longitudinal research as mental models evolve.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 5 Not valid · 20 Not checked · 195 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For RQ2, we found that participants' disclosure intentions were primarily affected by the perceived capability of the AI.”
plainstates a causen = 19- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Additionally, most participants did not know they could opt out of having ChatGPT use their data for model training.”
plainn = 19- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In this work, we provide the first set of empirical evidence supporting the speculation that the human-like text-generation capability of LLMs induces more sensitive disclosure behaviors in certain contexts.”
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, we also found almost every participant took ad-hoc privacy-protective measures featuring different levels of convenience and utility cost.”
plainn = 19- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This narrowed down our dataset to 30K conversations containing PII.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
114 more claims
Create a free account to see all 220 claims and every source verdict.
- It may state two separate claims as one. 54
- Probably not a substantial claim. 61
- The quote only partly states this claim. 1
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“a highly anthropomorphized design of Replikas most likely lead to reduced privacy concerns of its users.”
From "I tell him everything that I do": An investigation of privacy and safety implications of AI companion usage (2025) · cites this paper as [15]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Users often rely on inaccurate mental models of LLM data flows and are vulnerable to dark patterns, which impedes their ability to grasp real privacy risks [22,47].”
From Understanding Users' Privacy Perceptions Towards LLM's RAG-based Memory (2025) · cites this paper as [47]
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Beyond avoiding overt toxicity, many deployments require agents to comply with specific regulatory or policy constraints [6, 123].”
From Evaluation and Benchmarking of LLM Agents: A Survey (2025) · cites this paper as [123]
Experimental — accuracy not yet measuredIn the citing paper
“This, combined with human-like interactions and reasoning abilities that encourage sensitive disclosures, further increases user vulnerability to privacy-, cybersecurity-, and physical risks [15], [16], [18], [19].”
From "I tell him everything that I do": An investigation of privacy and safety implications of AI companion usage (2025) · cites this paper as [15]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ CCS CONCEPTS · p. 2
“we conducted semi-structured interviews with ChatGPT users (N=19)”
§ 4 INTERVIEW METHODOLOGY · p. 6
“we conducted semi-structured interviews with 19 users of LLM-based CAs (18 ChatGPT users, 1 Bing chat user).”
Citations
Cites
How Many Interviews Are Enough?: An Experiment with Data Saturation and Variability
Guest, G., Bunce, A., & Johnson, L. (2006) · Field Methods · doi:10.1177/1525822x05279903 · Cited by 20,808
The Coding Manual for Qualitative Researchers
Lungu, M. (2022) · American Journal of Qualitative Research · doi:10.29333/ajqr/12085 · Cited by 10,826 · cited 2×
“Our analysis involved two rounds of coding as recommended by Saldaña [61].”
MethodSocial Exchange Theory: An Interdisciplinary Review
Cropanzano, R., & Mitchell, M. S. (2005) · Journal of Management · doi:10.1177/0149206305279602 · Cited by 10,068 · cited 1×
“Prior research has also utilized theoretical frameworks like contextual integrity [52, 53] and social exchange theory [12] to analyze online disclosure behaviors.”
Training language models to follow instructions with human feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · arXiv · doi:10.48550/arxiv.2203.02155 · Cited by 4,208 · cited 1×
“” (P6)”
Training Language Models to Follow Instructions with Human Feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-2011 · Cited by 3,117
Privacy as Contextual IntegrityPreprint
Nissenbaum, H. F. (2004) · SSRN Electronic Journal · Cited by 2,891
A Survey of Large Language Models
Zhao, W. X., Zhou, K., Li, J., Tang, T., Dong, Z., Hou, Y., Zhang, B., Min, Y., Zhang, J., Liu, P., et al. (2026) · Frontiers of Computer Science · doi:10.1007/s11704-026-60308-3 · Cited by 2,093
McDonald, N., Schoenebeck, S., & Forte, A. (2019) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3359174 · Cited by 1,206
Grimm, P. (2010) · Wiley International Encyclopedia of Marketing · doi:10.1002/9781444316568.wiem02057 · Cited by 1,054
Privacy in context: technology, policy, and the integrity of social life
(2010) · Choice Reviews Online · doi:10.5860/choice.47-6940 · Cited by 928
Research works printed but not linked: 29
- Lukas Bieringer, Kathrin Grosse, Michael Backes, Battista Biggio, and Katharina Krombholz. 2022. Industrial practitioners’ mental models of adversarial machine learning. In Eighteenth Symposium on Usable Privacy and Security (SOUPS 2022). 97–116.
- Nicholas Carlini, Daphne Ippolito, Matthew Jagielski, Katherine Lee, Florian Tramer, and Chiyuan Zhang. 2022. Quantifying memorization across neural language models. arXiv preprint arXiv:2202.07646 (2022). · arXiv
- Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E Gonzalez, 2023. Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality. See https://vicuna. lmsys. org (accessed 14 April 2023) (2023).
- Catherine Dwyer, Starr Hiltz, and Katia Passerini. 2007. Trust and privacy concern within social networking sites: A comparison of Facebook and MySpace. AMCIS 2007 proceedings (2007), 339.
- Andrea Fox. 2023. ChatGPT scored 72study shows. https://www.healthcareitnews.com/news/chatgpt-scored-72-clinical-decision-accuracy-mgb-study-shows Accessed: 09/11/2023.
- Kevin Gallagher, Sameer Patil, and Nasir Memon. 2017. New Me: Understanding Expert and { Non-Expert} Perceptions and Usage of the Tor Anonymity Network. In Thirteenth Symposium on Usable Privacy and Security (SOUPS 2017). 385–398.
- Xinyang Geng, Arnav Gudibande, Hao Liu, Eric Wallace, Pieter Abbeel, Sergey Levine, and Dawn Song. 2023. Koala: A dialogue model for academic research. Blog post, April 1 (2023).
- Frances Grodzinsky and Herman T Tavani. 2010. Applying the “contextual integrity” model of privacy to personal blogs in the blogoshere. (2010).
- Arnav Gudibande, Eric Wallace, Charlie Snell, Xinyang Geng, Hao Liu, Pieter Abbeel, Sergey Levine, and Dawn Song. 2023. The False Promise of Imitating Proprietary LLMs. arXiv preprint arXiv:2305.15717 (2023). · arXiv
- Eszter Hargittai and Alice Marwick. 2016. “What can I really do?” Explaining the privacy paradox with online apathy. International journal of communication 10 (2016), 21.
- Joel Jang, Dongkeun Yoon, Sohee Yang, Sungmin Cha, Moontae Lee, Lajanugen Logeswaran, and Minjoon Seo. 2022. Knowledge unlearning for mitigating privacy risks in language models. arXiv preprint arXiv:2210.01504 (2022). · arXiv
- Nikhil Kandpal, Eric Wallace, and Colin Raffel. 2022. Deduplicating training data mitigates privacy risks in language models. In International Conference on Machine Learning. PMLR, 10697–10707.
- Ruogu Kang, Laura Dabbish, Nathaniel Fruchter, and Sara Kiesler. 2015. { “My} Data Just Goes { Everywhere:”} User Mental Models of the Internet and Implications for Privacy and Security. In Eleventh symposium on usable privacy and security (SOUPS 2015). 39–52.
- 10.18653/v1 · doi:10.18653/v1
- Xuechen Li, Florian Tramer, Percy Liang, and Tatsunori Hashimoto. 2021. Large language models can be strong differentially private learners. arXiv preprint arXiv:2110.05679 (2021). · arXiv
- 10.18653/v1 · doi:10.18653/v1
- Alexandra Mai, Katharina Pfeffer, Matthias Gusenbauer, Edgar Weippl, and Katharina Krombholz. 2020. User mental models of cryptocurrency systems-a grounded theory approach. In Sixteenth Symposium on Usable Privacy and Security (SOUPS 2020). 341–358.
- Jimit Majmudar, Christophe Dupuy, Charith Peris, Sami Smaili, Rahul Gupta, and Richard Zemel. 2022. Differentially private decoding in large language models. arXiv preprint arXiv:2205.13621 (2022). · arXiv
- KR McKee, X Bai, and S Fiske. 2021. Understanding human impressions of artificial intelligence. Preprint]. PsyArXiv. https://doi. org/10.31234/osf. io/5ursp (2021). · doi:10.31234/osf
- Yao Mu, Qinglong Zhang, Mengkang Hu, Wenhai Wang, Mingyu Ding, Jun Jin, Bin Wang, Jifeng Dai, Yu Qiao, and Ping Luo. 2023. EmbodiedGPT: Vision-Language Pre-Training via Embodied Chain of Thought. arXiv preprint arXiv:2305.15021 (2023). · arXiv
- Donald A Norman. 2014. Some observations on mental models. In Mental models. Psychology Press, 15–22.
- Heleen Rutjes, Martijn Willemsen, and Wijnand IJsselsteijn. 2019. Considerations on explainable AI and users’ mental models. In CHI 2019 Workshop: Where is the Human? Bridging the Gap Between AI and HCI. Association for Computing Machinery, Inc.
- Blase Ur, Sean M Segreti, Lujo Bauer, Nicolas Christin, Lorrie Faith Cranor, Saranga Komanduri, Darya Kurilova, Michelle L Mazurek, William Melicher, and Richard Shay. 2015. Measuring { Real-World} Accuracies and Biases in Modeling Password Guessability. In 24th USENIX Security Symposium (USENIX Security 15). 463–481.
- Da Yu, Saurabh Naik, Arturs Backurs, Sivakanth Gopi, Huseyin A Inan, Gautam Kamath, Janardhan Kulkarni, Yin Tat Lee, Andre Manoel, Lukas Wutschitz, 2021. Differentially private fine-tuning of language models. arXiv preprint arXiv:2110.06500 (2021). · arXiv
- Eric Zeng, Shrirang Mare, and Franziska Roesner. 2017. End user security and privacy concerns with smart homes. In thirteenth symposium on usable privacy and security (SOUPS 2017). 65–80.
- Chiyuan Zhang, Daphne Ippolito, Katherine Lee, Matthew Jagielski, Florian Tramèr, and Nicholas Carlini. 2021. Counterfactual memorization in neural language models. arXiv preprint arXiv:2112.12938 (2021). · arXiv
- 10.18653/v1 · doi:10.18653/v1
- Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, 2023. A survey of large language models. arXiv preprint arXiv:2303.18223 (2023). · arXiv
- Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, 2023. Judging LLM-as-a-judge with MT-Bench and Chatbot Arena. arXiv preprint arXiv:2306.05685 (2023). · arXiv
Cited by
Zhang, R., Li, H., Meng, H., Zhan, J., Gan, H., & Lee, Y.-C. (2025) · Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3706598.3713429 · Cited by 123
“Empirical evidence further suggests that interactions with AI companions can reduce loneliness, alleviate emotional distress, uplift moods, and foster self-reflection [23, 53, 61, 88].”
"I look at it as the king of knowledge": How Blind People Use and Understand Generative AI Tools
Adnin, R., & Das, M. (2024) · The 26th International ACM SIGACCESS Conference on Computers and Accessibility · doi:10.1145/3663548.3675631 · Cited by 73
“An important way to mitigate these harms is to improve people’s mental models of GenAI [98] so that they can question its capabilities and limitations and accordingly decide when and how to use these tools [55].”
BackgroundInfluentialEvaluation and Benchmarking of LLM Agents: A Survey
Mohammadi, M., Li, Y., Lo, J., & Yip, W. (2025) · Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 · doi:10.1145/3711896.3736570 · Cited by 74
The Emerged Security and Privacy of LLM Agent: A Survey with Case Studies
He, F., Zhu, T., Ye, D., Liu, B., Zhou, W., & Yu, P. S. (2024) · arXiv · doi:10.48550/arxiv.2407.19354 · Cited by 60
The Emerged Security and Privacy of LLM Agent: A Survey with Case Studies
He, F., Zhu, T., Ye, D., Liu, B., Zhou, W., & Yu, P. S. (2026) · ACM Computing Surveys · doi:10.1145/3773080 · Cited by 58
“[113] proposes a meta-framework for private deep learning that captures key principles from recent fine-tuning methods to enhance privacy without compromising performance.”
Metadata
- Authors
- Zhang, Z., Jia, M., Lee, H.-P. (., Yao, B., Das, S., Lerner, A., Wang, D., & Li, T.
- Affiliations
- Northeastern University (US)
- Carnegie Mellon University (US)
- Rensselaer Polytechnic Institute (US)
- ORCID
- Venue
- Proceedings of the CHI Conference on Human Factors in Computing Systems
- Publisher
- ACM
- Published
- 2024-05-11
- Pages
- 1-26
- OpenAlex
- W4396832466
- Cited by
- 95
- References
- 35
- Citations by year
- 2026: 43 · 2025: 45 · 2024: 7
- FWCI
- 124.09
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- CHI Conference on Human Factors in Computing Systems (publishedVersion, cc-by-sa)
- arXiv (Cornell University) (submittedVersion, cc-by)
- arXiv (Cornell University) (cc-by)
- Topics
- Social Sciences › Sociology and Political Science › Privacy, Security, and Data Protection
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Human-Computer Interaction › Innovative Human-Technology Interaction
- Keywords
- privacy disclosure, sensitive information disclosure, user privacy awareness, privacy-utility trade-off, dark patterns, user mental models, ChatGPT, human-AI interaction, ethical AI
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-10-30, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-30 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: “It's a Fair Game”, or Is It? Examining How Users Navigate Disclosure Risks and Benefits When Using LLM-Based Conversational Agents. https://app.cochat.ai/papers/10.1145/3613904.3642385
- The paper
- Zhang, Z., Jia, M., Lee, H.-P. (., Yao, B., Das, S., Lerner, A., Wang, D., & Li, T. (2024). “It's a Fair Game”, or Is It? Examining How Users Navigate Disclosure Risks and Benefits When Using LLM-Based Conversational Agents. Proceedings of the CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3613904.3642385
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.