Good-Enough Privacy in Platform Governance: Evidence from Fujian and Busan–Gyeongnam

1. Introduction

Artificial intelligence (AI) platforms have diffused rapidly into public-service delivery across East Asia, bringing efficiency gains alongside persistent privacy anxiety [[1[3]. Users acknowledge risk yet continue to engage—an apparent “privacy paradox” reported across domains [4,5,21,28,29]. Individual-level explanations (e.g., cognitive bias, risk–benefit miscalibration, habituation) clarify micro-level tolerance [6,21,28,29], but they understate how platform-governance arrangements set the boundaries within which people balance caution and compliance on digital services [20,23,30].

In highly institutionalized settings such as China and Korea, privacy choices are mediated by governance norms—visibility of explanations and accountability channels, and the reliability of operating procedures—that shape what counts as “good-enough” protection [20,23,30]. Building on bounded rationality and satisficing, we frame continued use as bounded compliance: engagement persists once minimum transparency and reliability thresholds are perceived, even when concern remains [14,17,21].This study conducts a symmetric, subnational survey in Fujian, China (N=185) and Busan–Gyeongnam, Korea (N=187) to examine how privacy concern and accountability visibility jointly sustain participation in AI-enabled public-service platforms. Using cross-regional heat maps and binary-logit average marginal effects, we show that willingness to use remains consistently high (≥0.70) across most conditions, while sensitivity patterns differ in ways that reflect institutional pathways of trust—with accountability visibility and institutional continuity acting as practical levers [8,20,23].

Contributions

(1)

We reframe the privacy paradox as an institutional satisficing problem in platform governance rather than an individual inconsistency [14,17,21].

(2)

We operationalize a symmetric diagnostic survey that isolates accountability and concern and discloses sparse subcells for transparency (n<10), while focusing inference on adequately populated cells [11,15].

(3)

We identify a “good-enough” compliance zone, showing how accountability visibility (resolution proofs, traceable updates) and institutional continuity (reliable operations, O&M capacity) stabilize trust thresholds on AI-enabled public-service platforms [20,23,30]. The remainder of the paper presents the theoretical grounding, data and measures, empirical results, and policy implications for trustworthy AI platforms [[1[3,27].

Figure 1. (a) Conceptual Framework.Notes. This framework links platform governance to empirical diagnostics and policy interpretation. It outlines: (i) institutional thresholds of transparency and reliability; (ii) symmetric surveys in Fujian (N = 185) and Busan–Gyeongnam (N = 187); (iii) diagnostic analyses using heat maps and logistic regression; and (iv) policy interpretation focused on feedback visibility and institutional continuity. Subcells with n < 10 are shown for transparency but excluded from statistical inference [11,15]. (b) Bounded Compliance Map: Institutional Pathways and Behavioral Convergence. Notes. This concept map illustrates relationships among key factors sustaining participation on AI-enabled public-service platforms. Three antecedents—privacy concern (Q6, blue), accountability visibility (Q7, green), and institutional continuity (green)—jointly shape perceived adequacy (amber), representing the “good-enough” threshold of transparency and reliability. When this threshold is reached, users enter the bounded-compliance zone (navy), a satisficing equilibrium where engagement persists despite concern, leading to continued participation (Q8, red; willingness ≥ 0.70). Regional pathways (purple) differ—Korea shows accountability sensitivity (+12.5 pp), China shows continuity assurance—yet both converge toward similar behavioral outcomes.

Figure 1. (a) Conceptual Framework.Notes. This framework links platform governance to empirical diagnostics and policy interpretation. It outlines: (i) institutional thresholds of transparency and reliability; (ii) symmetric surveys in Fujian (N = 185) and Busan–Gyeongnam (N = 187); (iii) diagnostic analyses using heat maps and logistic regression; and (iv) policy interpretation focused on feedback visibility and institutional continuity. Subcells with n < 10 are shown for transparency but excluded from statistical inference [11,15]. (b) Bounded Compliance Map: Institutional Pathways and Behavioral Convergence. Notes. This concept map illustrates relationships among key factors sustaining participation on AI-enabled public-service platforms. Three antecedents—privacy concern (Q6, blue), accountability visibility (Q7, green), and institutional continuity (green)—jointly shape perceived adequacy (amber), representing the “good-enough” threshold of transparency and reliability. When this threshold is reached, users enter the bounded-compliance zone (navy), a satisficing equilibrium where engagement persists despite concern, leading to continued participation (Q8, red; willingness ≥ 0.70). Regional pathways (purple) differ—Korea shows accountability sensitivity (+12.5 pp), China shows continuity assurance—yet both converge toward similar behavioral outcomes.

2. Literature Review and Theoretical Grounding

Privacy paradox beyond psychology.

A large body of research explains why people keep using data-intensive services despite perceived risk—via cognitive dissonance, risk–benefit miscalibration, trust heuristics, or habitual normalization [4,5,6,21,28,29]. These accounts illuminate micro-level tolerance, yet they often understate the institutional settings that delimit what users regard as acceptable exposure and reassurance in digital environments [20,23,30].

Platform governance as the missing layer.

In highly institutionalized contexts, platform governance—norms and routines around transparency, accountability, and operational reliability—mediates privacy decisions [1,2,20,23]. Transparency and feedback visibility can signal procedural fairness and responsiveness, bolstering trust in government/platform institutions [8,20]. Accountability channels (complaint handling, audit trails, explanation mechanisms) help users perceive protections as “good enough,” shifting the problem from a psychological paradox to an institutional assurance challenge [23,30]. Hence, the privacy paradox should be read as a function of embedded governance mechanisms rather than individual inconsistency alone [27,30].

Satisficing and bounded compliance.

Building on bounded rationality [14,17], continued use is conceptualized as satisficing: users settle for adequate outcomes under informational and institutional constraints. Where accountability signals are visible and responsive, the perceived adequacy threshold is higher; where reassurance stems from stable but less dialogic routines, engagement persists once minimal reliability is perceived [6,14,17,20]. We frame this as bounded compliance—behavior as pragmatic adaptation rather than inconsistency—consistent with findings in automation trust and risk-adjusted choice models [8,9,23]. Crucially, adequacy operates here as a moderating condition (governance cues → adequacy threshold → sustained participation), not as a mediating construct that explains variance ex post.

Relation to prior satisficing-equilibrium work.

This study extends the satisficing-equilibrium logic developed in prior work on multi-actor trust in AI-enabled smart tourism by the same authors [31]. Whereas that framework theorizes SE at a broader systems level, the present analysis narrows to subnational, platform-governance thresholds (accountability visibility; institutional continuity) and examines how they sustain bounded compliance under privacy concern. The two studies are complementary: here we operationalize platform-level adequacy and test symmetric, diagnostic expectations without causal claims.

Comparative implication for East-Asian platforms.

Applied to AI-enabled public-service platforms in East Asia, the framework anticipates similar behavioral outcomes (continued use amid concern) realized through two institutional pathways [1,3,10,20,23]:

• ▪ Accountability-visibility pathway: clear explanations, traceable resolution proofs, participatory feedback;
• ▪ Institutional-continuity pathway: reliable operations, predictable O&M routines, service stability.

Prior comparative work in digital governance and smart services supports these divergent yet convergent logics of trust formation [19,20,23,27]. Our design therefore examines how concern and accountability jointly condition willingness to use, without presuming a universal psychological mechanism [4,5,21].

Resulting analytical model.

Synthesizing the above, we propose a bounded-compliance model of platform governance in which privacy perceptions (concern) translate into behavioral participation (use) conditional on institutional cues—particularly accountability visibility and system reliability [14,17,20,23,30]. The model yields diagnostic expectations for threshold-based trust on AI platforms and motivates the symmetric China–Korea survey that follows [11,15].

Analytical expectations.

• E1. High baseline willingness (≥ 0.70) is expected to persist across most conditions when either accountability visibility or reliability reaches perceived adequacy.
• E2. Korea is expected to show stronger sensitivity to accountability visibility, while China tends to rely more on institutional continuity.
• E3. Under high concern, willingness converges to a “good-enough” zone once at least one governance lever signals adequacy.

3. Methods and Data

Design

We implement a symmetric, diagnostic survey comparing Fujian, China (N = 185) and Busan–Gyeongnam, Korea (N = 187) to examine how privacy concern (Q6) and accountability/explanation (Q7) jointly shape willingness to use AI services (Q8) on public-service platforms. The design emphasizes cross-regional parity, nonparametric visualization, and transparent handling of sparse subcells, following best practices in comparative governance research [12,19,20]. Subcells with n < 10 are disclosed for transparency and treated as diagnostic rather than inferential, in line with ethical small-sample reporting [11,15].

Measures and coding.

We analyze three binary indicators (coded to {0,1}):

• Q6 – Privacy concern: 0 = low, 1 = high.
• Q7 – Accountability/explanation: 0 = absent, 1 = present.
• Q8 – Willingness to use AI services: 0 = low, 1 = high.

For each region, we compute the 2×2 conditional probability table of Pr(HighUse = 1 | Q6, Q7) and a difference panel to visualize cross-regional contrasts. Δ is defined cell-by-cell as BG minus FJ (in percentage points). These summaries support diagnostic comparability and bounded-compliance checks across subsamples [11,15,17]. Descriptive statistics appear in Table 1.

Visualization and estimation.

We present dual-color heat maps for each region and a Δ panel (Figure 2). For proportions and small cells, we report 95% confidence intervals using Wilson’s method, which is preferable to normal approximations in this setting [11,15]. Where subcell counts are sparse (e.g., n < 10), estimates are labeled diagnostic and excluded from formal inference.

Measures. Q6 (concern), Q7 (accountability), Q8 (use). Cells display Pr(HighUse = 1 | Q6, Q7) × 100; the right panel shows Δ (percentage points). Estimation. Ninety-five percent CIs via Wilson’s method; scales harmonized for legibility.

Reading. Both regions sustain high probabilities (≥ 0.70). The largest Δ (+12.5 pp) appears in low-concern + high-accountability, indicating stronger accountability sensitivity in Korea, while Fujian patterns are consistent with implicit institutional assurance—a motif noted in East-Asian digital-trust literature [1,3,20,23]. Sparse subcells (e.g., BG: Concern = 0 and Account = 1, n = 5) are reported for transparency only [11,15].

Robustness checks.

We assess stability through diagnostic resampling and sensitivity checks for each region’s 2×2 table and the Δ panel. Figure 3 overlays Fujian (left), Busan–Gyeongnam (middle), and Δ = BG − FJ (right) on common scales. Across these checks, cell probabilities remain high (typically ≥ 0.70), and the sign of Δ aligns with the main heat maps, confirming stability and directional robustness. Full descriptive summaries and probability ranges appear in Table 2 and the supplementary CSV files.

Bivariate tests and regression (for completeness).

We also report (i) region–variable chi-square tests (Table 3A) and (ii) a logistic specification with Q6, Q7, their interaction, and Region as covariates (Table 3B). These outputs are read as pattern-consistent diagnostics, not hypothesis tests, and are included solely to align with common reporting practice in socio-behavioral diagnostics [15].

Ethics and Data Availability. This study analyzes anonymized adult survey data (N = 372) approved under protocol YSUIRB-202511-HR-194-02 at Youngsan University. The dataset represents an adult-only subset of previously authorized minimal-risk social-behavioral research and involves no intervention or collection of personally identifiable information [12,19]. No new data collection was conducted for this analysis.

Data availability. De-identified cell matrices and Δ values (CSV) are provided as Supplementary Materials, together with variable labels and coding keys to support replication. The governance-cue coding (Q5–Q7) follows the Information Control Level (ICL) structure reported in the related satisficing-equilibrium study [31]. As the composite is formative rather than reflective, internal-consistency coefficients are not used as validity diagnostics; details of bilingual wording and coding can be provided upon request.

Multicollinearity was assessed in the Satisficing Equilibrium (SE) dataset [31], where all variance inflation factors (VIF ≈ 1.03–1.10) indicated distinct, non-interchangeable governance cues among Q5–Q7. Given that the present study draws on the same measurement structure with a reduced diagnostic specification, further VIF tests were unnecessary.

The coding of governance cues (Q5–Q7) follows the Information Control Level (ICL) structure previously validated in the satisficing-equilibrium study [31]. Each item was measured on a 0–3 ordinal scale reflecting ascending adequacy of transparency, concern (reverse-coded), and accountability. The composite is formative rather than reflective, so internal-consistency coefficients are not diagnostic (Cronbach’s α ≈ 0.29, VIF ≈ 1.03–1.10 [31]). The full bilingual wording and detailed coding scheme are available upon request.

4. Discussion

To situate our findings, Table 4 contrasts this study with influential work on the privacy paradox, trust, and governance. Psychological and calculus-based explanations emphasize individual trade-offs under risk and uncertainty [21,28], while governance-centered studies show that transparency can lift institutional trust yet rarely specify platform-level accountability and reliability levers [20]. Human–automation/HCI research clarifies when reliance is “appropriate” but typically abstracts from subnational governance variation and public-service platforms [8,23].

Our symmetric China–Korea survey adds three elements. First, we recast continued use as bounded compliance—a satisficing equilibrium that arises once accountability visibility (clear explanations, traceable resolution) and operational continuity (predictable O&M) cross “good-enough” adequacy thresholds. We treat adequacy as a moderating condition, not a mediating construct. Second, we operationalize these platform thresholds with diagnostic cell probabilities and bootstrap checks, aligning measurement with governance cues rather than individual psychometrics. Third, we show pathway differences—Korea is more accountability-sensitive, whereas China relies more on implicit institutional assurance—yet both converge to sustained participation even under high concern. This shifts the emphasis from a paradox located in the individual toward threshold governance on platforms.

Policy implications. For AI-enabled public-service platforms, two levers are actionable:

• ▪ Feedback visibility. Publish explanation artifacts and closure proofs (e.g., ticket IDs, time-to-resolution, status traces) at the user interface to make accountability verifiable.
• ▪ Continuity capacity. Harden O&M routines (uptime SLAs, incident-drill coverage, recovery-time targets) and communicate them plainly.

Calibrating these levers to local expectations can raise perceived adequacy and sustain engagement even under elevated concern, without over-promising universal psychological change.

5. Interpretive Synthesis and Conclusions

The symmetric survey comparison between Fujian (N = 185) and Busan–Gyeongnam (N = 187) reveals a consistent behavioral mechanism on AI-enabled public-service platforms: users continue to participate despite privacy concern once minimal transparency and reliability thresholds are perceived [20,23]. Heat-map and bootstrap diagnostics (Figure 2 and Figure 3) indicate high willingness to use (≥ 0.70) across concern–accountability conditions, with stronger accountability sensitivity in Korea and stronger institutional assurance in China. We interpret this pattern as bounded compliance—a satisficing equilibrium structured by institutional cues rather than a contradiction in individual psychology [14,17,21,28].

Regional pathways.

⯎

Fujian. Engagement aligns with institutional continuity (predictable procedures, stable O&M), which reassures even under concern—consistent with evidence that reliability and continuity calibrate reliance on automated or platformized services [8,23].

⯎

Busan–Gyeongnam. Engagement aligns with accountability visibility (explanations, feedback closure), which raises the perceived “good-enough” protection threshold, echoing findings that transparency cues lift institutional trust [20].

Despite these pathway differences, both regions converge on sustained participation under bounded conditions—consistent with satisficing under bounded rationality [14,17] and with privacy-paradox regularities observed in prior work [21,28,29].

Conclusions.

(1)

Privacy satisficing. Continued participation stems not from optimizing privacy but from achieving a good-enough trust threshold once safeguards are perceived as credible [14,17,21].

(2)

Governance-shaped thresholds. Where visibility and redress channels are salient, users expect higher procedural openness; where continuity and reliability dominate, concern is tolerated if operations remain predictable [8,20,23].

(3)

Diagnostic value of symmetry. Treating sparse subcells as diagnostic and maintaining regional parity enables transparent comparison of platform trust without overstated inference, aligning with best practice for small-cell analysis [11,15].

Policy implications.

⮚

China (Fujian). Reinforce feedback visibility—publish closure evidence and service-level commitments—while sustaining multi-year O&M continuity to stabilize satisficing thresholds [8,20].

⮚

Korea (Busan–Gyeongnam). Ensure institutional continuity within accountability systems as AI services scale; procedural volatility could erode trust [20,23].

⮚

Both contexts. Develop platform-trust dashboards tracking two levers—accountability visibility and operational continuity—as leading indicators of bounded compliance [20,23].

Limitations and future research. The design is diagnostic: some cross-tab cells are sparse and reported for transparency rather than inference [11,15]. The analysis is cross-sectional and limited to two subnational cases. Future work should extend to additional regions, employ longitudinal tracking, and experimentally vary visibility and continuity to observe shifts in satisficing thresholds [8,20,23].

Overall synthesis. Viewing the privacy paradox as governance-dependent bounded compliance offers a grounded explanation for why caution and participation coexist on public-service platforms. Institutions that calibrate transparency and reliability—rather than attempting to eliminate concern—can sustain trust and foster responsible AI adoption across diverse governance contexts [14,17,20,21,23,28,29].