Big Data attract a lot of researchers because of the techniques and approaches that it offers for treating huge volumes of information circulating on the Internet. The objective of this work is to study of the problem security in Big Data. We present some of the methods and techniques used in Big Data security which exist in reality, and based on the criteria that we find in those work we proposed an approach based on the multi-agent system (SMA) to protect the safety of the information in Big Data. In this approach we used seven agent, and they are divided as follows: four for external security independent from Hadoop, and three for internal security. Our approach treat the problems of integrity, authentication, privacy and access control, intrusion detection. To To give some significant results, we used conventional tools of information systems (MySql, NoSql....etc) as well as new tools for handling large volumes of data such as Hadoop, biservice-ce, and Pentaho etc.