Security, privacy, poor data quality and opaque AI systems undermine managerial decision-making and erode business sustainability; firms with weaker sustainability strategies suffer the most.

Main Finding

AI systems’ “dark sides” — security, privacy, quality, and lack-of-transparency threats — increase the incidence of poor decision-making quality, and poor decision-making quality, in turn, reduces business sustainability. A weak (vs. proactive) sustainability strategy amplifies the negative effect of poor decision-making quality on business sustainability. The paper frames these relationships through socio-technical systems (STS) theory.

Key Points

• Definitions and scope
  • Dark sides of AI systems are defined as four practical, deployment-level threats: security (e.g., adversarial attacks, data poisoning), privacy (e.g., data leaks, re-identification), quality (e.g., poor data, bias, reliability failures), and lack of transparency (black-box decisions).
  • Poor decision-making quality captures faulty framing, unreliable information, unclear values/trade-offs, and defective reasoning resulting from AI-system failures or misuse.
  • Business sustainability (BS) covers environmental, social, and operational dimensions of long-term firm viability.
• Conceptual model
  • Dark-side threats → poor decision-making quality (positive relationship).
  • Poor decision-making quality → lower business sustainability (negative relationship).
  • Proactive Sustainability Strategy (PSS) moderates the latter link: weaker PSS strengthens the negative impact.
• Theoretical framing
  • Socio-technical systems (STS) theory is used: AI systems are interdependent technical and social subsystems; misalignment produces the observed harms.
• Managerial takeaways highlighted by authors
  • Addressing AI security, privacy, transparency, and quality is necessary not only for operational performance but for sustaining long-term environmental, social, and reputational outcomes.
  • Robust AI governance and joint optimisation of social and technical subsystems are recommended to manage value–destruction risks.

Data & Methods

• Design: Mixed-methods study.
  • Qualitative component: used to operationalise measurement items (item development and construct definition).
  • Quantitative component: survey-based empirical analysis.
• Sample: 310 respondents from diversified industries (energy & utilities; manufacturing & industrial; agriculture, food & beverage; technology & electronics; transportation & logistics; hospitality & tourism; consumer goods/FMCG). Respondents included board members, chief sustainability officers, senior leadership, and employees.
• Analysis: Statistical testing of hypothesised relationships (paper reports empirical support for the proposed paths and moderation). Grounding in STS theory.
• Publication metadata: Int. J. Business Information Systems (Vol. 52, No. 6, 2026); DOI 10.1504/IJBIS.2026.10079894. Open Access (CC BY).
• Notes on limitations (implied): cross-sectional survey limits causal claims; industry and regional coverage may constrain generalisability; reliance on self-reported perceptions for constructs such as threats and sustainability.

Implications for AI Economics

• Firm-level investment and risk pricing
  • Costs of AI adoption should include investments in security, privacy protections, transparency (explainability), and data-quality assurance. Failure to internalise these costs can produce negative externalities that reduce long-run firm value.
  • Risk-adjusted ROI: firms must evaluate AI projects using risk-adjusted returns that reflect potential sustainability damages (reputational, regulatory, litigation) from the dark sides.
• Market competition and adoption dynamics
  • Short-term productivity gains from AI may be offset by longer-term sustainability losses if dark-side risks are unaddressed; this can affect competitive dynamics, with firms that embed robust governance capturing durable advantage.
• Labour, productivity, and allocative efficiency
  • Poor AI-driven decisions can degrade workforce morale, increase turnover, and reduce human capital productivity — generating economic frictions beyond immediate operational loss.
• Regulatory and policy implications
  • Evidence supports policies that mandate disclosure, auditability, and minimum standards for AI security/privacy/explainability to correct market failures and informational asymmetries.
  • Public policy (regulation, certification, liability rules) will affect the cost of deploying AI and thus shape diffusion patterns and investment in safer AI.
• Financial-sector impacts
  • Insurance and credit markets may price premiums differently for firms with weak PSS or poor AI governance; sustainability strategy strength should be treated as a risk-mitigating signal.
• Research and measurement recommendations for AI economics
  • Incorporate socio-technical risk factors (security, privacy, transparency, data quality) into productivity and TFP studies of AI adoption.
  • Use longitudinal and transaction-level data to estimate causal impacts of AI dark-side incidents on firm value, employment, and environmental/social performance.
• Managerial policy: firms should internalise joint optimisation (STS): integrate technical fixes (robust models, secure pipelines, explainability tools) with social/organisational measures (governance, training, sustainability strategy) to protect both short-term performance and long-run sustainability.

If you want, I can (a) extract the paper’s hypothesised model diagram and present it as a simple schematic, (b) draft short policy recommendations for regulators and insurers based on these results, or (c) summarize the specific survey measures and statistical results if you provide the results section text.