Mobile botnets have proliferated with the popularization of mobile and portable devices, being a simple and powerful method to launch Distributed Denial of Service (DDoS) attacks. This letter presents a stochastic adaptive model for mobile botnets' dynamics and their self-organized and self-adaptive behavior to generate DDoS attacks. The bots' collaborations combine reinforcement and fading rules based upon the level of servers' activity and map to a time-varying weighted directed graph. This model can explain the natural emergence of two distinct time-scales when bots massively attack a server.