Large Language Models (LLMs) have revolutionized Generative Artificial Intelligence (GenAI) tasks, becoming an integral part of various applications in society, including text generation, translation, summarization, a...Large Language Models (LLMs) have revolutionized Generative Artificial Intelligence (GenAI) tasks, becoming an integral part of various applications in society, including text generation, translation, summarization, and more. However, their widespread usage emphasizes the critical need to enhance their security posture to ensure the integrity and reliability of their outputs and minimize harmful effects. Prompt injections and training data poisoning attacks are two of the most prominent vulnerabilities in LLMs, which could potentially lead to unpredictable and undesirable behaviors, such as biased outputs, misinformation propagation, and even malicious content generation. The Common Vulnerability Scoring System (CVSS) framework provides a standardized approach to capturing the principal characteristics of vulnerabilities, facilitating a deeper understanding of their severity within the security and AI communities. By extending the current CVSS framework, we generate scores for these vulnerabilities such that organizations can prioritize mitigation efforts, allocate resources effectively, and implement targeted security measures to defend against potential risks.展开更多
Currently,cybersecurity threats such as data breaches and phishing have been on the rise due to the many differentattack strategies of cyber attackers,significantly increasing risks to individuals and organizations.Tr...Currently,cybersecurity threats such as data breaches and phishing have been on the rise due to the many differentattack strategies of cyber attackers,significantly increasing risks to individuals and organizations.Traditionalsecurity technologies such as intrusion detection have been developed to respond to these cyber threats.Recently,advanced integrated cybersecurity that incorporates Artificial Intelligence has been the focus.In this paper,wepropose a response strategy using a reinforcement-learning-based cyber-attack-defense simulation tool to addresscontinuously evolving cyber threats.Additionally,we have implemented an effective reinforcement-learning-basedcyber-attack scenario using Cyber Battle Simulation,which is a cyber-attack-defense simulator.This scenarioinvolves important security components such as node value,cost,firewalls,and services.Furthermore,we applieda new vulnerability assessment method based on the Common Vulnerability Scoring System.This approach candesign an optimal attack strategy by considering the importance of attack goals,which helps in developing moreeffective response strategies.These attack strategies are evaluated by comparing their performance using a variety ofReinforcement Learning methods.The experimental results show that RL models demonstrate improved learningperformance with the proposed attack strategy compared to the original strategies.In particular,the success rateof the Advantage Actor-Critic-based attack strategy improved by 5.04 percentage points,reaching 10.17%,whichrepresents an impressive 98.24%increase over the original scenario.Consequently,the proposed method canenhance security and risk management capabilities in cyber environments,improving the efficiency of securitymanagement and significantly contributing to the development of security systems.展开更多
Any computer system with known vulnerabilities can be presented using attack graphs. An attacker generally has a mission to reach a goal state that he expects to achieve. Expected Path Length (EPL) [1] in the context ...Any computer system with known vulnerabilities can be presented using attack graphs. An attacker generally has a mission to reach a goal state that he expects to achieve. Expected Path Length (EPL) [1] in the context of an attack graph describes the length or number of steps that the attacker has to take in achieving the goal state. However, EPL varies and it is based on the “state of vulnerabilities” [2] [3] in a given computer system. Any vulnerability throughout its life cycle passes through several stages that we identify as “states of the vulnerability life cycle” [2] [3]. In our previous studies we have developed mathematical models using Markovian theory to estimate the probability of a given vulnerability being in a particular state of its life cycle. There, we have considered a typical model of a computer network system with two computers subject to three vulnerabilities, and developed a method driven by an algorithm to estimate the EPL of this network system as a function of time. This approach is important because it allows us to monitor a computer system during the process of being exploited. Proposed non-homogeneous model in this study estimates the behavior of the EPL as a function of time and therefore act as an index of the risk associated with the network system getting exploited.展开更多
文摘Large Language Models (LLMs) have revolutionized Generative Artificial Intelligence (GenAI) tasks, becoming an integral part of various applications in society, including text generation, translation, summarization, and more. However, their widespread usage emphasizes the critical need to enhance their security posture to ensure the integrity and reliability of their outputs and minimize harmful effects. Prompt injections and training data poisoning attacks are two of the most prominent vulnerabilities in LLMs, which could potentially lead to unpredictable and undesirable behaviors, such as biased outputs, misinformation propagation, and even malicious content generation. The Common Vulnerability Scoring System (CVSS) framework provides a standardized approach to capturing the principal characteristics of vulnerabilities, facilitating a deeper understanding of their severity within the security and AI communities. By extending the current CVSS framework, we generate scores for these vulnerabilities such that organizations can prioritize mitigation efforts, allocate resources effectively, and implement targeted security measures to defend against potential risks.
基金supported by the Institute of Information&Communications Technology Planning&Evaluation(IITP)grant funded by the Korea Government(MSIT)(No.RS2022-II220961).
文摘Currently,cybersecurity threats such as data breaches and phishing have been on the rise due to the many differentattack strategies of cyber attackers,significantly increasing risks to individuals and organizations.Traditionalsecurity technologies such as intrusion detection have been developed to respond to these cyber threats.Recently,advanced integrated cybersecurity that incorporates Artificial Intelligence has been the focus.In this paper,wepropose a response strategy using a reinforcement-learning-based cyber-attack-defense simulation tool to addresscontinuously evolving cyber threats.Additionally,we have implemented an effective reinforcement-learning-basedcyber-attack scenario using Cyber Battle Simulation,which is a cyber-attack-defense simulator.This scenarioinvolves important security components such as node value,cost,firewalls,and services.Furthermore,we applieda new vulnerability assessment method based on the Common Vulnerability Scoring System.This approach candesign an optimal attack strategy by considering the importance of attack goals,which helps in developing moreeffective response strategies.These attack strategies are evaluated by comparing their performance using a variety ofReinforcement Learning methods.The experimental results show that RL models demonstrate improved learningperformance with the proposed attack strategy compared to the original strategies.In particular,the success rateof the Advantage Actor-Critic-based attack strategy improved by 5.04 percentage points,reaching 10.17%,whichrepresents an impressive 98.24%increase over the original scenario.Consequently,the proposed method canenhance security and risk management capabilities in cyber environments,improving the efficiency of securitymanagement and significantly contributing to the development of security systems.
文摘Any computer system with known vulnerabilities can be presented using attack graphs. An attacker generally has a mission to reach a goal state that he expects to achieve. Expected Path Length (EPL) [1] in the context of an attack graph describes the length or number of steps that the attacker has to take in achieving the goal state. However, EPL varies and it is based on the “state of vulnerabilities” [2] [3] in a given computer system. Any vulnerability throughout its life cycle passes through several stages that we identify as “states of the vulnerability life cycle” [2] [3]. In our previous studies we have developed mathematical models using Markovian theory to estimate the probability of a given vulnerability being in a particular state of its life cycle. There, we have considered a typical model of a computer network system with two computers subject to three vulnerabilities, and developed a method driven by an algorithm to estimate the EPL of this network system as a function of time. This approach is important because it allows us to monitor a computer system during the process of being exploited. Proposed non-homogeneous model in this study estimates the behavior of the EPL as a function of time and therefore act as an index of the risk associated with the network system getting exploited.
