Exploits are widely used in different attack scenarios. To support efficient threat detection and security risk management, they have to be timely detected and assessed. The task of automated objective and unbiased assessment of the exploits is still not solved. This paper proposes an original approach to the automated assessment of the exploits using deep learning methods. It is based on the assumption that an exploit’s severity depends on the exploited vulnerability and the exploit source code itself. The proposed approach incorporates a data gathering and pre-processing technique, a data processing technique, and a trained deep learning model that defines the exploit class considering its severity for information security. The Exploit-DB database was used as a training set, the exploits were labeled using metrics from the Common Vulnerability Scoring System for the related vulnerabilities. The CodeBERT-based model was trained to perform classification of the exploits, the experiments showed that the trained model demonstrates accuracy equal to 0.93 on all data gathered for the experiments.