Balancing efficiency and responsibility in generative AI adoption

Purpose This study investigates how firms achieve sustainable outcomes from generative artificial intelligence (GenAI) adoption by balancing efficiency gains with ethical and environmental responsibilities. It integrates dynamic capability theory and responsible AI governance to examine the mediating role of digital dynamic capability (DsDC) and the moderating effect of responsible AI orientation (RAIO). Design/methodology/approach Two complementary empirical studies were undertaken. Study 1 applies a staggered difference-in-differences design with propensity-score matching (DiD-PSM) to 500 GenAI adoption events (2019–2024). Study 2 employs a two-step system generalised method of moments (GMM) estimator for 1,274 firm-year observations (2016–2024) to analyse persistence, mediation and moderation effects. Findings GenAI adoption was associated with a market-perceived efficiency signal (a 2.7% rise in cumulative abnormal returns) alongside a 1.9% decline in environmental, social and governance sentiment, a reputational rather than substantive sustainability indicator, evidencing an efficiency-responsibility tension. Over time, DDC helped partially translate these short-term gains into sustainable performance, while RAIO mitigated ethical and reputational risks, enhancing firms’ long-term legitimacy. Research limitations/implications The study relies on archival indicators and text-derived governance measures; future research could combine these data with surveys or internal audit evidence. Practical implications Managers should pair GenAI deployment with adaptive digital capability and responsible AI governance to protect stakeholder trust. Originality/value This study advances theory by integrating capability and governance perspectives to explain the sustainability of GenAI adoption. Methodologically, it introduces temporal triangulation by combining DiD-PSM and System GMM, offering a novel dynamic causal framework for responsible digital transformation.