When many rivals cut corners, firms are likelier to launch unethical AI; a single extremely unethical competitor deters launch, but prevention-focused corporate objectives can blunt that deterrent and sustain responsible choices.

Main Finding

Competitive context shapes firms’ decisions to deploy unethical AI in distinct ways: widespread (horizontal) unethical competition increases firms’ likelihood of launching unethical AI regardless of managerial regulatory focus, while the presence of a single highly unethical competitor (vertical competition) also pushes firms toward unethical deployment — but prevention-focused organizational goals can mitigate that vertical effect. Moral disengagement processes help explain these behavioral shifts.

Key Points

• Two types of unethical competition were distinguished:
  • Horizontal: many competitors using similar unethical AI tactics.
  • Vertical: a single competitor engaging in especially severe unethical AI.
• Horizontal unethical competition reliably increases the launch/adoption of unethical AI services (a “race to the bottom”), and this effect is insensitive to managers’ regulatory focus.
• Vertical unethical competition (severity of a competitor’s unethical behavior) tends to increase firms’ adoption of unethical AI (i.e., reduce responsible-AI deployment).
• Prevention-focused organizational goals (emphasizing avoiding harm/losses) can counteract the effect of a severe unethical competitor and promote more responsible AI decisions.
• Study 3 provides evidence that moral disengagement is an underlying mechanism linking competitive context to RAI choices.

Data & Methods

• Overall approach: three online experimental studies using scenario-based vignettes where participants evaluated the decision to launch an AI-based digital service described as unethical.
• Manipulations:
  • Horizontal unethical competition: presence vs absence (many competitors doing the same unethical tactics).
  • Vertical unethical competition: severity level of a single competitor’s unethical behavior.
  • Organizational goal / regulatory focus:
    • Study 1 (N = 249): measured participants’ regulatory focus (chronic promotion vs prevention tendencies).
    • Study 2 (N = 304): manipulated organizational goals to induce promotion- or prevention-focused decision framing.
  • Study 3 (N = 159): tested moral disengagement as a mediating psychological mechanism.
• Key dependent variable: willingness/decision to launch the unethical AI service (operationalized in scenario judgments).
• Main statistical patterns:
  • Horizontal competition → higher propensity to launch unethical AI (robust across studies).
  • Vertical (severe) competitor → higher propensity to launch unethical AI (i.e., reduced RAI deployment), moderated by organizational prevention focus such that prevention framing reduced the effect.
  • Measures in Study 3 implicate moral disengagement (e.g., justifications, diffusion of responsibility) as mediating the relationship between competitive pressures and unethical choices.

Implications for AI Economics

• Market dynamics and externalities:
  • Competition can create negative externalities in AI markets: horizontal competition leads to a coordination failure (race-to-the-bottom) where firms adopt unethical practices to stay competitive.
  • Vertical shocks (an extreme unethical actor) can propagate unethical norms unless firms’ internal goals check them.
• Policy and regulation:
  • Regulatory interventions (minimum safety/ethics standards, penalties, disclosure requirements, audits) are needed to correct market failures and reduce incentives to adopt unethical AI.
  • Policies that raise the reputational/regulatory cost of unethical deployment (e.g., mandatory impact reporting, liability rules) would blunt both horizontal and vertical competitive pressures.
• Firm strategy and governance:
  • Internal governance matters: embedding prevention-focused objectives (risk- and harm-avoidance incentives, compliance-oriented KPIs, ethical review gates) can reduce the likelihood of responding to competitive pressure by lowering moral disengagement.
  • Firms and industry groups can mitigate race-to-the-bottom dynamics via standards, certifications, and voluntary coalitions that change payoff structures.
• Modeling recommendations for AI economics research:
  • Models of firm behavior should include distinct competition modalities (horizontal vs vertical unethical actions), organizational goal heterogeneity (promotion vs prevention focus), and behavioral channels (moral disengagement, reputation dynamics).
  • Welfare analyses should incorporate dynamic strategic interactions where ethical choices have externalities on regulators, consumer trust, and adoption.

Limitations & next steps (brief): - Experimental vignette design increases internal control but limits ecological generalizability; field and archival studies are needed. - The “unethical AI” treatments are scenario-based and may vary by industry, cultural context, and actor type; heterogeneity of firms, consumer reactions, and enforcement regimes should be explored.