This study proposes an autonomous early decision system for cyber security related contents of Twitter. In the context, both cyber and non-cyber security related tweets are collected and the obtained data is trained by means of Naive Bayes Classifier. Besides, Term Frequency - Inverse Document Frequency (TF-IDF) term weighting method is used for vectorization purpose. Experimental results show that, the developed system can classify the tweets in terms of their cyber security related or non-related security with the 70.03% success rate. It can be included that the system can be used as an alert system on Twitter for early cyber-attack detection.