<p>This paper investigates a privacy-preserving leaderless consensus problem for a class of nonlinear multiagent systems (MASs) under attacks. For false data injection attacks, by a finite-time attack detector for nonlinear MASs with privacy-preserving signals, a distributed detection-isolation algorithm is developed to detect and isolate compromised agents within nonlinear dynamics. Via an amplification technique for signals in an improved Liu cryptosystem, the sensitivity of the information is preserved, and decryption errors between plaintext encryption and decryption are mitigated, ensuring both privacy preservation and satisfactory recovery of plaintext information. With privacy-preserving and attack-free information, a privacy-preserving leaderless consensus control strategy is then developed via backstepping and reinforcement learning (RL) techniques. This privacy-preserving RL-based consensus control strategy compensates for unknown dynamics and errors between true signals and decrypted ones with fewer learning parameters. With graph theory and Lyapunov stability theory, it is also proven that the output consensus error and all signals are ultimately bounded. Simulation examples are provided to validate the effectiveness of this control strategy.</p>