CoChat

The DOI matches its Crossref record; 57 of the 57 references it cites were found in a registry or an index; 7 of the 46 sources its claims cite were checked for support.

Bridging risk and retention: risk coping with privacy and hallucination in AI

Open access
Shin, Don, Rasul, Azmat, Shahzad, Khuram · 2026 · AI & SOCIETY

Abstract

This study examines how users engage with generative AI systems in the presence of perceived technological risks, focusing on the interplay between risk perception, adaptive behavior, and continued use. Drawing on a risk coping perspective, the study conceptualizes generative AI use as an ongoing process in which users evaluate potential informational and privacy-related threats and adjust their practices accordingly. the analysis explores how perceived risks shape user attitudes, coping strategies, and patterns of sustained engagement. Rather than treating risk as a simple deterrent, the findings show that users often respond through adaptive behaviors such as information verification and privacy management, enabling continued use despite recognized limitations. By distinguishing between different forms of perceived risk and user responses, the study advances understanding of how individuals navigate uncertainty in AI-mediated environments. More broadly, the findings illuminate shifting expectations around responsibility, trust, and epistemic labor in everyday interactions with generative AI technologies, with implications for human–AI interaction, digital risk governance, and the sustainability of AI-supported knowledge practices.

Overview

CoChat Summary

TL;DR

This study applies risk coping theory (RCT) to examine how perceived hallucination and privacy risks relate to attitudes, protective behaviors, and continuance intention toward generative AI. It finds risks are linked to less favorable attitudes but also to adaptive verification and privacy-protective behaviors, with information verification moderating the attitude–continuance relationship.

Key findings

  • Privacy concerns were negatively associated with attitudes toward GAI (β = -0.19, p < 0.001) and positively associated with personal-level (β = 0.38, p < 0.001) and system-level (β = 0.37, p < 0.001) privacy-protective behaviors.
  • Perceived hallucination risk was negatively associated with attitudes toward GAI (β = -0.23, p < 0.001) and positively associated with information verification behaviors (β = 0.18, p < 0.001).
  • Attitudes positively predicted continuance intention (β = 0.38, p < 0.001); information verification was positively associated with continuance (β = 0.21, p = 0.001) and negatively moderated the attitude–continuance relationship (β = -0.24, p = 0.003), weakening it at higher verification levels.
  • Neither personal nor system-level privacy-protective behaviors significantly predicted continuance intention (β = 0.01, p = 0.63; β = -0.03, p = 0.47), and their interactions with attitudes were nonsignificant.

Methods

  • Cross-sectional, self-report study of individual-level perceptions and behaviors; RCT threat appraisal was operationalized as perceived hallucination risk and perceived privacy risk, with protective responses as information verification and privacy-protective actions. The main sample size is not stated in the provided text; scale development included qualitative interviews with 10 frequent GAI users and a pilot test (N = 102).
  • Measures: privacy concern (3 items adapted from Ioannou & Tussyadiah 2021); perceived hallucination risk (4 items adapted from Hannigan et al. 2023); information verification (3 items adapted from Flanagin & Metzger 2000); privacy-protective behaviors (7 items developed, covering personal- and system-level dimensions); attitude toward GAI (4 semantic differential items); continuance intention (3 items adapted from Liu & Huang 2024).
  • Analysis: covariance-based structural equation modeling in Mplus 8 with maximum likelihood estimation; measurement model fit was χ²/df = 2.91, CFI = 0.97, TLI = 0.96, RMSEA = 0.05 [0.038, 0.060], SRMR = 0.06; reliability and validity thresholds were met, and Harman’s single-factor test showed the first factor accounted for 28.2% of variance.

Limitations

  • Stated: the analysis is centered on individual-level perceptions, attitudes, and self-reported behaviors; broader systemic risks (institutional data governance, training data opacity, platform design, regulatory shortcomings) are not directly measured, so findings should not be read as an assessment of structural or institutional risk mitigation.
  • Stated: the cross-sectional design limits assessment of how risk perceptions, coping strategies, and usage intentions evolve over time; self-reported measures may be subject to common method bias and social desirability effects; the model excludes self-efficacy, perceived agency, digital literacy, and emerging system-level hallucination-reduction features.
  • Stated/evident: the study is a partial application of RCT—coping appraisal components such as self-efficacy and response efficacy were not directly modeled; only factuality hallucinations were examined, not faithfulness hallucinations; the main sample size and recruitment details are not stated in the provided text.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1007/s00146-026-03127-1

Cites

57 of 57 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

7 Valid · 5 Not valid · 36 Not checked · 71 Own results and interpretations

55 more claims

Create a free account to see all 119 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 43
  • It may state two separate claims as one. 28
  • Borderline: one of the tests for a substantial claim is unclear. 1

Create a free account to see these claims.

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.

4 of 4

Citations

Cites

Show all 57 references

Research works printed but not linked: 1

  1. Aiken LS, West SG (1991) Multiple regression. Sage Publications, Newbury Park, CA

Metadata

Authors
Shin, D., Rasul, A., & Shahzad, K.
Affiliations
  • Zayed University (AE)
  • University of the Punjab (PK)
Venue
AI & SOCIETY
Published
2026-08-04
Volume
41
Issue
9
Pages
8407-8419
ISSN
  • 0951-5666 (ISSN-L)
  • 1435-5655
OpenAlex
W7172386007
Cited by
0
References
43
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
AI & Society (publishedVersion, cc-by)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
  • Social Sciences › Decision Sciences › Information Systems and Management › Technology Adoption and User Behaviour
  • Social Sciences › Sociology and Political Science › Privacy, Security, and Data Protection
Keywords
generative AI, risk perception, risk coping, privacy, hallucinations, continued use, adaptive behavior, fact-checking, privacy management, trust in AI, human-AI interaction, user attitudes

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

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: Bridging risk and retention: risk coping with privacy and hallucination in AI. https://app.cochat.ai/papers/10.1007/s00146-026-03127-1
The paper
Shin, D., Rasul, A., & Shahzad, K. (2026). Bridging risk and retention: risk coping with privacy and hallucination in AI. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03127-1

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.