Levine et al.’s resource-rational contractualism (RRC) models moral judgment as a cost-sensitive approximation to ideal bargaining among all affected parties. By allowing agents to alternate among agreement-based, welfare-based, and action-based shortcuts according to resource demands, the framework attempts to explain much of the observed variability in moral reasoning. What remains less explored is how agents can lower their moral costs by increasing the complexity of the interdependent decision context itself. We examine three potential strategies: strategic delegation (shifting moral choice to another agent), diffusion of responsibility (blurring causal responsibility and weakening action standards’ effectiveness), and coalitional boundary-setting (narrowing the stakeholder set to in-group members). These reduce moral costs by altering or bypassing virtual bargaining. We sketch extensions to RRC that introduce explicit parameters for agency transfer, credit assignment, and stakeholder inclusion, bringing the model into closer alignment with the social settings in which many moral decisions are made.

RRC presupposes that individual agents approximate a virtual bargaining process in which they simulate mutually acceptable policies by integrating the (anticipated) preferences of all stakeholders. However, in practice, individuals and organizations often circumvent this requirement by strategically delegating morally sensitive decisions to others. Research documents how individuals use intermediaries when decisions may entail morally questionable outcomes, allowing them to distance themselves from responsibility or reputational harm (Bartling & Fischbacher, Reference Bartling and Fischbacher2012; Erat, Reference Erat2013; Gawn & Innes, Reference Gawn and Innes2019; Weiss & Forstmann, Reference Weiss and Forstmann2024, Reference Weiss and Forstmann2025).

This highlights an unmodeled RRC route for bypassing virtual bargaining. Delegation reshapes the moral decision by distributing both gains and possible penalties across multiple actors. When the decision maker/principal delegates a decision to an equal who will similarly benefit, their personal motivation to run a full simulation of all stakeholders’ interests weakens for two reasons. First, by relocating responsibility to the delegate, the principal may assume – often self-servingly – that the other party will be accounting for the welfare of those who might be harmed.

Second, social evaluation becomes softer for the principal: observers can be expected to focus primary blame on the delegate who carried out the act, leaving the initiator exposed to only a reduced share of criticism, loss of status, or formal sanction (Coffman, Reference Coffman2011; Paharia et al., Reference Paharia, Kassam, Greene and Bazerman2009). As personal penalties diminish while cognitive costs rise (modeling delegate incentives, others’ perceptions, and so on), virtual bargaining becomes an unattractive use of mental resources. Thereby, delegation – to a subordinate or a peer – can reduce the motivation to engage in the kind of thorough mental simulation of others’ preferences that RRC assumes underlies moral judgment. To capture this dynamic, we propose that RRC be extended with a cost-of-agency-transfer parameter that reflects both the motivational gains of offloading cognitive effort and the perceived reduction in personal accountability. Agents weigh deliberation benefits against both cognitive costs and delegation advantages (i.e., against reduced effort/blame minus potential outsourcing costs). Full virtual bargaining is then undertaken only when its expected benefits exceed both the raw cognitive cost of simulation and the net savings from delegating.

Under RRC, agents undertake effortful virtual bargaining whenever the anticipated costs from moral failure exceed the cognitive cost of simulating all affected parties. When decisions involve multiple agents, individual social costs decrease while modeling costs rise. As a result, agents should more readily default to simple heuristics – most notably, action standards.

These standards, however, seem to presuppose a decision context in which an individual a) cannot outsource moral decision-making to others and b) can clearly track their own causal relevance. In multi-agent contexts (e.g., bystander crowds (Fischer et al., Reference Fischer, Krueger, Greitemeyer, Vogrincic, Kastenmüller, Frey and Kainbacher2011), collaborative transgressions (Irlenbusch et al., Reference Irlenbusch, Mussweiler, Saxler, Shalvi and Weiss2020; Keshmirian et al., Reference Keshmirian, Hemmatian, Bahrami, Deroy and Cushman2022), however, agents are affected by two interconnected processes that promote diffusion of responsibility and lower the effectiveness of action standards.

First, agents may (strategically) use others’ actions as social cues for action standards or bargaining outcomes. When bystanders remain passive, for example, this inaction can be interpreted as evidence that intervention violates prevailing norms or reflects tacit outcomes of others’ virtual bargaining. Second, agents must track their causal pivotality – their assessment of whether their specific (in-)action would decisively change the outcome, which affects potential sanctions after moral failure.

We therefore propose extending RRC with a responsibility-tracking parameter that reflects perceived pivotality via counterfactual thinking (Lagnado, Gerstenberg, & Zultan, Reference Lagnado, Gerstenberg and Zultan2013). This parameter scales the expected moral or reputational cost of (non-)action by assessing the likelihood that the agent’s action or inaction would have changed the outcome. When personal/individual responsibility is high, agents override inferred group standards and activate salient action standards. When responsibility is low, the benefit of intervention/cheating remains below/above the threshold, and agents continue to defer to the group’s apparent consensus.

While delegation and diffusion blur agent-outcome connections, individuals and groups can go further by narrowing the set of people whose interests enter the calculation. RRC assumes that agents begin with a representation of all morally relevant stakeholders and then approximate ideal contractualist reasoning using computationally efficient strategies. In many real-world contexts, however, this sequence is reversed. Agents first form a coalition and tacitly restrict the scope of moral consideration to its members. Subsequent approximations apply only within that reduced frame, with outsiders removed from the equation.

This restriction has systematic consequences and may explain various moral disagreements. Action standards, such as rules about fairness or procedural compliance, are interpreted as binding within the group (e.g., “treat everyone equally”) while remaining inapplicable beyond it. The appeal of this strategy is both cognitive and strategic: It is less demanding to represent a small, like-minded group than a diverse set of affected parties, and shared responsibility for group-based standards dilutes reputational consequences.

To account for these dynamics, we propose extending RRC with a cost-of-inclusion parameter that reflects the informational, emotional, and reputational burdens of incorporating additional stakeholders. When inclusion costs (gathering preferences, negotiating, risking disagreement) exceed exclusion costs (sanctions, reputational damage, guilt), agents will rationally limit the scope of moral reasoning to their coalition. This extension preserves the logic of RRC while allowing it to explain how immoral cooperation within multi-agent contexts can be achieved through the systematic exclusion of outsiders from virtual bargaining.