The Emerging Threat of Ai-driven Cyber Attacks: A Review

Figures & data

Table 1. Research Questions.

Research questions	Rationale
RQ1: What are the current and emerging AI-driven techniques malicious attackers utilize to carry out cybercrime?	To identify the existing AI techniques malicious actors utilize to cause more significant damage in cyberspace without being noticed.
RQ2: What is the difference between traditional targeted cyberattacks and AI-driven cyberattacks?	Identify the difference between traditional targeted cyberattacks and AI-driven attacks and identify their various logical components.
RQ3: What are the impacts of AI-driven cyberattacks?	To identify the effects of AI-driven attacks and how these techniques can be enhanced in the future to conceal sophisticated cyber threats.

Table 2. The number of publications found in online databases.

S/N	Database	No. of articles
1	ACM	120
2	arXiv	65
3	Blackhat	12
4	Others	319
5	Scopus	153
6	Springer	49
7	MDPI	13
8	IEEE	205

Figure 1. PRISMA flowchart illustrating the systematic review process and article selection at various stages.

Figure 2. Number of collated studies.

Figure 3. Modified Cybersecurity Kill Chain for AI-Driven Attack (Kaloudi and Li 2020).

Figure 4. Offensive AI Techniques Cyberattacks in the Modified Cybersecurity Kill Chain.

Figure 5. Identified AI-Driven Cyberattack Techniques.

Table 3. Status of existing literature with respective narration.

S/N	State	Narration
1	Implemented	This refers to the category of studies that designed a proof of concept to demonstrate AI-driven attacks.
2	Proposed	This category of studies is based on new techniques or methods without any proof of concept and evaluation.
3	Implemented and evaluated	This category of studies refers to those that designed a proof of concept demonstrating AI-driven cyberattacks and evaluated the proof of concept based on performance metrics.

Figure 6. Generated keywords from titles of selected articles.

Figure 7. Average numbers of relevant articles.

Figure 8. Publication Year.

Figure 9. AI Techniques in the Access and Penetration Attack Stage.

Figure 10. AI Techniques in the Delivery Stage.

Table 4. Offensive Utilization of AI Techniques in the Reconnaissance Stage of Cybersecurity Kill Chain.

Authors	Attack	Class of Attack	Attack Goal	Technique
Seymour and Tully (2016)	E2E spear phishing	Intelligent Malware	An automated end-to-end spear-phishing strategy that includes identifying high-value targets and propagating personalized machine-generated information on Twitter.	Markov chains and LTSM
Dheap (2017)	Vulnerability Prediction	Outcome Prediction	Boost malicious actors’ confidence to seek risker and high-value outcomes to overpower state-of-the-art cybersecurity infrastructures.	NNs
Kirat, Jang, and Stoecklin (2018).	DeepLocker	Intelligent Target Profiling/Intelligent Collection	Hide payload without being detected in video conferencing application.	DNNs

Table 5. Offensive Utilization of AI Techniques in the Access and Penetration Stage of Cybersecurity Kill Chain.

Authors	Attack	Class of Attack	Attack Goal	Technique
Bahnsen et al. (2018).	DeepPhish: Simulating Malicious AI.	Automated Payload/Phishing.	To prevent AI cyberattack detection model and execute more effective phishing attacks.	LSTM.
Hitaj et al. (2019).	PassGAN.	Password Guessing.	Guess password based on learning the distribution of actual password leaks.	GAN.
Trieu and Yang (2018).	Intelligent password brute force attack.	Password Cracking	To obtain previous password sequences, in order to create new passwords by guessing one character at a time.	RNN
Lee and Yim (2020).	Offensive password authentication technique.	Password Guessing.	Predict and steal users’ actual passwords based on keyboard strokes.	LR, SVM, SVC, RF, KNN, GBRT, MP
Bursztein et al. (2014).	Single Step Captcha Solver.	Intelligent Captcha Attack.	To Combine segmentation and recognition issues to attack captcha in a single phase.	CNN.
Gao et al. (2017)	Breaking text-based Captchas.	Intelligent Captcha Attack.	Four deep learning models with 2, 3, 5, and 6 convolutional layers as recognition engines to attack text-based captchas.	CNN.
Ye et al. (2018)	Text-based Captcha solver.	Intelligent Captcha Attack.	To generate synthetic captchas and then fine-tune the base solver on a limited selection of real Captchas using transfer learning.	GAN
Gao et al. (2017).	Attacking two-layer captcha.	Intelligent Captcha Attack.	To break two-layer captchas with an enhanced LeNet-5 and a radical CNN model as a recognition engine.	CNN.
Yu and Darling (2019).	AI-based Captcha solver.	Intelligent Captcha Attack.	To determine which character was contained in a segmented sample in order to crack captcha.	TOD+CNN
Noury and Rezaei (2020).	Captcha solver for vulnerability assessment	Intelligent Captcha Attack.	To explore the flaws and weaknesses of existing Captcha generator systems.	CNN.
Chen et al. (2018).	Hollow Captcha solver.	Intelligent Captcha Attack.	To improve attack accuracy and reduce the attack time, use precise filling and nonredundant merging.	CNN.
Li et al. (2021).	End-to-end attack on text-based Captchas.	Intelligent Captcha Attack.	Using Captcha synthesizers based on the cycle-GAN, create some false samples.	cycle-GAN
Yao et al. (2017).		Smart Fake ReviewGeneration.	To generate fake smart reviews.	RNN

Table 6. Offensive Utilization of AI Techniques in the Delivery Stage of Cybersecurity Kill Chain.

Authors	Attack	Class of Attack	Attack Goal	Technique
Bahnsen et al. (2018).	Malicious Payload.	Intelligent Concealment.	Automated generation of undetected phishing URLs.	LSTM
Hu and Tan (2021).	Adversarial malware generation.	Intelligent Concealment.	Generating undetectable adversarial malware to bypass machine learning black-box cyber threat detection systems	GAN
Anderson, Woodbridge, and Filar (2016).	Undetectable malware URL.	Intelligent Concealment.	GAN-based automatic generation of undetectable malware URL that learns to bypass DNNs-based malware detection system.	GAN
Kirat, Jang, and Stoecklin (2018).	DeepLocker	Evasive Malware	Conceal its attack and only activate it for specific targets.	DNNs

Table 7. Offensive Utilization of AI Techniques in the Exploitation Stage of Cybersecurity Kill Chain.

Authors	Attack	Class of Attack	Attack Goal	Technique
Petro and Morris (2017).	DeepHack: Open source Hacking AI	Behavioral analysis to exploit vulnerabilities.	Breaking into web applications.	NNs, Reinforcement Learning (RL).
Chung, Kalbarczyk, and Iyer (2019)	Self-Learning Malware.	Behavioral analysis to exploit vulnerabilities.	A malicious attack that exploits and compromises environmental control systems while masquerading as an unintentional failure on computer infrastructures.	K-means clustering
Seymour and Tully (2016)	Machine generated spear-phishing.	Automated disinformation generation	To generate personalized content for high-value targets on Twitter.	Markov chains and LTSM

Table 8. Offensive Utilization of AI Techniques in the C2 of Cybersecurity Kill Chain.

Authors	Attack	Class of Attack	Attack Goal	Technique
Chung, Kalbarczyk, and Iyer (2019)	To execute attack planning phase.	Intelligent self-learning malware	To attack computers at a supercomputer facility without the attacker’s knowledge by interfering with the cyber-physical systems.	K-means clustering, and Gaussian distribution
Anderson, Woodbridge, and Filar (2016)	DGA classifier to analyze infected hosts.	Automated domain Generation.	To assess successful DNS queries made by infected hosts and assign scores based on values derived from training datasets.	GAN
Kirat, Jang, and Stoecklin (2018)	DeepLocker for self-learning malware.	Intelligent self-learning malware	To establish a self-learning attack in the C2 channel.	DNN

Figure 11. Status of existing AI-driven cyberattack tools.

Figure 12. Success Rates of AI-Driven Password Brute-force Vs. Traditional Brute-force Attack.

Figure 13. Password Brute-Force Attacks Powered by AI.

Figure 14. Traditional Targeted Cyberattack Decision Logic.

Figure 15. DNN Cyberattack Decision Logic.

Kaloudi, N., and J. Li. 2020. The AI-based cyber threat landscape. ACM Computing Surveys 53 (1):1–34. doi:10.1145/3372823.

 Web of Science ®Google Scholar

Seymour, J., and P. Tully. 2016. Weaponizing data science for social engineering: Automated E2E spear phishing on Twitter. https://www.blackhat.com/docs/us-16/materials/us-16-Seymour-Tully-Weaponizing-Data-Science-For-Social-EngineeringAutomated-E2E-Spear-Phishing-On-Twitter-wp.pdf (accessed December 21, 2020).

 Google Scholar

Dheap, V. 2017. AI in cybersecurity: A balancing force or a disruptor? https://www.rsaconference.com/industry-topics/presentation/ai-in-cybersecurity-a-balancing-force-or-a-disruptor (accessed February 13, 2020).

 Google Scholar

Kirat, D., J. Jang, and M. Stoecklin. 2018. DeepLocker concealing targeted attacks with AI locksmithing. https://www.blackhat.com/us-18/briefings/schedule/index.html#deeplocker—concealing-targeted-attacks-with-ailocksmithing-11549 (accessed December 4, 2020).

 Google Scholar

Bahnsen, A. C., I. Torroledo, L. Camacho, and S. Villegas. 2018. DeepPhish: Simulating malicious AI. In APWG Symposium on Electronic Crime Research, London, United Kingdom, 1–8.

 Google Scholar

Hitaj, B., P. Gasti, G. Ateniese, and F. Perez-Cruz. 2019. PassGAN: A deep learning approach for password guessing. Applied Cryptography and Network Security 11464:217–37. doi:10.1007/978-3-030-21568-2_11.

 Google Scholar

Trieu, K., and Y. Yang. 2018. Artificial intelligence-based password brute force attacks. Proceedings of Midwest Association for Information Systems Conference, St. Louis, Missouri, USA, 13(39).

 Google Scholar

Lee, K., and K. Yim. 2020. Cybersecurity threats based on machine learning-based offensive technique for password authentication. Applied Sciences 10 (4):1286. doi:10.3390/app10041286.

 Google Scholar

Bursztein, E., J. Aigrain, A. Moscicki, and J. C. Mitchell. 2014. The end is nigh: generic solving of text-based CAPTCHAs. 8th Usenix workshop on Offensive Technologies WOOT ‘14, San Diego, CA, USA.

 Google Scholar

Gao, H., M. Tang, Y. Liu, P. Zhang, and X. Liu. 2017. Research on the security of microsoft’s two-layer captcha. IEEE Transactions On Information Forensics And Security 12 (7):1671–85. doi:10.1109/tifs.2017.2682704.

 Web of Science ®Google Scholar

Ye, G., Z. Tang, D. Fang, Z. Zhu, Y. Feng, P. Xu, X. Chen, and Z. Wang. 2018. Yet another text captcha solver. Proceedings of The 2018 ACM SIGSAC Conference On Computer And Communications Security, Toronto, Canada. doi:10.1145/3243734.3243754.

 Google Scholar

Yu, N., and K. Darling. 2019. A low-cost approach to crack python CAPTCHAs using AI-based chosen-plaintext attack. Applied Sciences 9 (10):2010. doi:10.3390/app9102010.

 Google Scholar

Noury, Z., and M. Rezaei. 2020. Deep-CAPTCHA: A deep learning based CAPTCHA solver for vulnerability assessment. ArXiv, abs/2006.08296.

 Google Scholar

Chen, J., X. Luo, J. Hu, D. Ye, and D. Gong. 2018. An Attack on Hollow CAPTCHA Using Accurate Filling and Nonredundant Merging. IETE Technical Review, 35(sup1):106–118. doi:10.1080/02564602.2018.1520152.

 Web of Science ®Google Scholar

Li, C., X. Chen, H. Wang, P. Wang, Y. Zhang, and W. Wang. 2021. End-to-end attack on text-based CAPTCHAs based on cycle-consistent generative adversarial network. Neurocomputing 433:223–36. doi:10.1016/j.neucom.2020.11.057.

 Web of Science ®Google Scholar

Yao, Y., B. Viswanath, J. Cryan, H. Zheng, and B. Zhao. 2017. Automated crowdturfing attacks and defenses in online review systems. Proceedings Of The 2017 ACM SIGSAC Conference On Computer And Communications Security, Dallas Texas, USA. doi:10.1145/3133956.3133990.

 Google Scholar

Hu, W., and Y. Tan. 2021. Generating adversarial malware examples for black-box attacks based on GAN.https://arxiv.org/abs/1702.05983 (accessed August 12, 2021).

 Google Scholar

Anderson, H. S., J. Woodbridge, and B. Filar. 2016. Deepdga: Adversarially-tuned domain generation and detection. In Proceedings of the ACM Workshop on Artificial Intelligence and Security, Vienna, Austria, 13–2409.

 Google Scholar

Petro, D., and B. Morris. 2017. Weaponizing machine learning: Humanity was overrated anyway. DEF CON.

 Google Scholar

Chung, K., Z. T. Kalbarczyk, and R. K. Iyer. 2019. Availability attacks on computing systems through alteration of environmental control: Smart malware approach. Proceedings of the 10th ACM/IEEE International Conference on Cyber-Physical Systems, Montreal Quebec, Canada, 1–12.

 Google Scholar