Paper Image

Using pre-trained language models for network security

Published on:

23 August 2023

Primary Category:

Cryptography and Security

Paper Authors:

Maria Rigaki,

Ondřej Lukáš,

Carlos A. Catania,

Sebastian Garcia

Bullets

Key Details

Pre-trained language models can act as proficient cybersecurity agents

LLM agents solve simulated security environments without extra training

LLM agents outperform some reinforcement learning methods

A new customizable network security environment is presented

GPT-4 exceeds GPT-3.5 performance as a cybersecurity agent

AI generated summary

Using pre-trained language models for network security

This paper explores using large pre-trained language models as agents for network penetration testing and cybersecurity scenarios. The models are able to successfully navigate simulated environments and make sequential decisions without additional training, outperforming some reinforcement learning methods. A new modular cybersecurity environment is also introduced.

Answers from this paper

Comments

No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up