P2P networks and distributed systems have become the key enabling technologies for the most innovative services and applications, such as Blockchain or the dark web. Actually, P2P architectures show relevant advantages such as reliability, transparency, or accountability. Furthermore, traditional intruders and malware can infect and affect the global behavior of these networks with a very low probability (because of the large number of participating nodes). However, distributed systems are forced to communicate and transfer data between P2P nodes. And those data are vulnerable, as they can be captured while being sent. Nowadays, intelligent attacks are especially dangerous and worrying, as illegitimate learning models capture private information to generate new knowledge and discover patterns. In standard protection mechanisms, encryption techniques are used to prevent access to data, but two main open questions arise. First, powerful new models can learn from encrypted data and, second, some information is always public and vulnerable, such as the Internet addresses. Therefore, new protection technologies for P2P networks are needed. In this paper, we propose a solution based on false synthetic information injection. The vulnerability and state of P2P nodes is described using a dynamical system, which analyzes how nodes change their exposure with time, so at any moment it is possible to estimate the percentage of exposed data and communication flows. With this estimation, a computational model based on elementary functions and describing the learning level of potential (and unknown) intelligent attackers is run, so it is possible to deduct the amount of false information to be injected, and intelligent attackers get confused, and their learning level reduces. An experimental validation supported by simulation scenarios is provided as well, and results prove the proposed scheme reduces the attacking success rate by up to 21%.