Although dishonesty is often a social phenomenon, it is primarily studied in individual settings. However, people frequently collaborate and engage in mutual dishonest acts. We report the first meta-analysis on collaborative dishonesty, analyzing 87,771 decisions (21 behavioral tasks; k = 123; n = 10,923). We provide an overview of all tasks used to measure collaborative dishonesty, and inform theory by conducting moderation analyses. Results reveal collaborative dishonesty is higher (i) when financial incentives are high; (ii) in lab than field studies; (iii) when third parties experience no negative consequences; (iv) in the absence of experimental deception; and (v) when groups consist of more males and (vi) younger individuals. Further, in repeated interactions, group members’ behavior is correlated —participants lie more when their partners lie—and lying increases as the task progresses. These findings are in line with the justified ethicality theoretical perspective, suggesting prosocial concerns increase collaborative dishonesty, whereas honest-image concerns attenuate it. We discuss how findings inform theory, setting an agenda for future research on the collaborative roots of dishonesty.