1

 

The perceived impact of Artificial Intelligence 

on cyber security operations in the South 

African banking sector 

 

Edward van Bosch 

 

2419308 

Supervisor: Dr Ayanda Magida 

 

 

Submitted to the Faculty of Commerce, Law and Management, University of 

the Witwatersrand, in fulfilment for the degree of Master of Management in the 

field of Digital Business 

 

 

 



 2

Declaration  

 

I, __Edward John van Bosch___, declare that this research report is my own work 

except as indicated in the references and acknowledgements. It is submitted in partial 

fulfilment of the requirements for the degree of Master of Management in the field of 

Digital Business at the University of the Witwatersrand, Johannesburg. It has not been 

submitted before for any degree or examination in this or any other university.  

 

Name: Edward John van Bosch   Signature: Edward John van Bosch  

 

Signed at …Randburg………………… 

 

On the ……28th……………. day of ……April………… 2025….  

  



 3

Abstract 

Cybersecurity threats are increasing exponentially, and South Africa is one of the top 

countries targeted by cyber criminals. Banks are adopting AI-powered cybersecurity 

solutions to fortify themselves against the ever-increasing cyber threat landscape, but 

this presents opportunities and challenges. To ascertain the efficacy and impact of AI 

in cybercrime eradication, the TOE framework and Perception Theory were adopted 

to cover the operating landscape of CS officers and guided the cognition and the 

rationalisation of AI adoption. The data from the literature review was posited against 

qualitative findings from semi-structured interviews of key CS personnel and was 

analysed using thematic analysis. The findings indicate a positive perception of AI as 

a cybersecurity force multiplier, particularly in enhancing real-time threat detection and 

automating response mechanisms. However, concerns were also raised about 

overreliance on AI, data privacy and the potential exploitation of vulnerabilities and 

algorithms within AI-powered cybersecurity solutions. The findings clearly 

demonstrated that AI is not a turnkey solution and does not possess the power and 

ability to eradicate cybercrime in totality but is the only solid viable defensive solution 

for cyber-attacks. AI cannot eradicate crime completely but is the most efficient tool 

and system in cybercrime detection and prevention. There was a marked disparity in 

perception of AI between direct AI users and literary conjecture on the impact of AI. 

Given the vastness of AI, a novel bespoke framework was created to guide new users 

on the necessity, risk, benefits and scale of AI implementation intra-organisationally.  

 

Keywords: AI, cybersecurity, cyberattacks, force multiplier, overreliance, solution. 



 4

Acknowledgements  

 

To my creator, without whom nothing is impossible, I praise you for giving me the 

opportunities, abilities and persistence to complete this research. 

 

To my wife Chantelle and daughters Kyla and Talor, thank you for always motivating 

and supporting me. We sacrificed a lot of time together. Thank you for being part of 

my life and your inspiration on this journey. 

 

To my friend Adelia: thank you for your encouragement, your belief in me, and your 

motivation through very tough and challenging times to see this through and to finish 

my studies and research.  You always inspired me to be more and to be better, and 

for that I will always be grateful. Thank you.   

 

Thank you to all the participants who participated and who took time away from their 

very busy schedules. 

 

Dr Ayanda Magida, thank you for guiding me through this process. 

To everyone who assisted me on this journey — there are so many of you. Thank you 

very much for your support. 

  



 5

Table of Contents 
Declaration ................................................................................................................ 2 

Abstract ..................................................................................................................... 3 

Acknowledgements .................................................................................................. 4 

List of Figures .......................................................................................................... 9 

List of Tables .......................................................................................................... 10 

List of Abbreviations .............................................................................................. 11 

Chapter 1: Introduction .......................................................................................... 13 

1.1 Statement of Purpose ............................................................................................. 13 

1.2 Background of the Study ........................................................................................ 14 

1.3 Research Problem ................................................................................................... 21 

1.4 Research Questions ................................................................................................ 21 

1.5 Justification for the study ....................................................................................... 22 

1.6 Delimitations of the Study ...................................................................................... 23 

1.7 Report Outline ......................................................................................................... 24 

Chapter 2: Literature Review ................................................................................. 26 

2.1 Introduction ............................................................................................................. 26 

2.2 Defining the TOE Framework ................................................................................. 28 

2.3 Technological component of TOE affecting CS operations ................................. 29 

2.3.1 The necessity of Cybersecurity ........................................................................................... 30 

2.3.2 AI CS Anthropomorphism .................................................................................................... 33 

2.4 Organisational component of TOE affecting CS operations ................................ 40 



 6

2.5 Environmental component of TOE affecting CS operations ................................ 42 

2.6 Challenges to AI CS operations ............................................................................. 43 

2.6.1 Technological Challenges .................................................................................................... 43 

2.6.2 Organisational Challenges .................................................................................................. 46 

2.6.3 Environmental Challenges .................................................................................................. 47 

2.7 Defining Perception Theory .................................................................................... 47 

2.7.2 Cognition, Science and Perception Flow ............................................................................ 51 

2.8 AI-CS and PT – A Synthesis .................................................................................... 52 

2.9 Literature Review Findings Viz Research Questions ............................................ 55 

2.10 Conclusion ............................................................................................................. 56 

Chapter 3: Research Methodology ....................................................................... 58 

3.1 Introduction ............................................................................................................. 58 

3.2 Research Philosophy .............................................................................................. 59 

3.3 Research Assumptions ........................................................................................... 61 

3.4 Research Design ..................................................................................................... 62 

3.5 Research Strategy and Method .............................................................................. 63 

3.6 Population................................................................................................................ 64 

3.7 Target Population and Sample Group .................................................................... 65 

3.8 Unit of Analysis ....................................................................................................... 66 

3.9 Data Collection ........................................................................................................ 67 

3.10 Data Analysis ......................................................................................................... 69 

3.11 Quality Assurance ................................................................................................. 72 

3.12 Ethical Considerations .......................................................................................... 73 



 7

Chapter 4: Findings and Discussion of Studies .................................................. 75 

4.1 Introduction ............................................................................................................. 75 

4.2 Summary Background of Participants ................................................................... 75 

4.3 Interview Themes and Rationale ............................................................................ 77 

4.4 Interview Findings ................................................................................................... 79 

4.4.1 Current use of AI in CS teams ............................................................................................. 79 

4.4.2 Benefits of AI ....................................................................................................................... 82 

4.4.3 Risks of AI in CS .................................................................................................................. 85 

4.4.4 CS-related Risk Mitigation Strategies ................................................................................. 90 

4.4.5 Impact on Security Posture ................................................................................................. 95 

4.4.6 Automation on Activities ...................................................................................................... 97 

4.4.7 AI’s Impact on CS Engineers ............................................................................................... 98 

4.4.8 Decision-Making Visibility .................................................................................................. 103 

4.4.9 Data Management and Governance ................................................................................. 107 

4.4.10 Protecting AI Systems ..................................................................................................... 111 

4.4.11 Protecting non-AI Systems .............................................................................................. 113 

4.4.12 Does AI give companies a competitive edge .................................................................. 113 

4.5 Conclusion............................................................................................................. 114 

Chapter 5: Conclusion & Recommendations .................................................... 116 

5.1 Introduction ........................................................................................................... 116 

5.2 Conclusions regarding Research Question 1 ..................................................... 117 

5.3 Conclusions regarding Research Question 2 ..................................................... 118 

5.4 Conclusions regarding Research Question 3 ..................................................... 119 

5.5 Limitations of this study ....................................................................................... 120 

5.6 Recommendations ................................................................................................ 121 



 8

5.7 Suggestions for further studies ........................................................................... 124 

References ............................................................................................................ 127 

APPENDIX A ......................................................................................................... 139 

APPENDIX B ......................................................................................................... 141 

APPENDIX C ......................................................................................................... 143 

APPENDIX D ......................................................................................................... 145 

 

  



 9

List of Figures  

Figure 1.1:Top three cyber risks in the financial services sector .............................. 17 

Figure 1.2: Percentage of Fin companies underinvesting on CS capabilities ........... 19 

Figure 2.1: Difference between Simple and Deep Learning Neural Networks .......... 37 

Figure 2.2: Four Types of Metaheuristic Algorithms in CS operations ...................... 38 

Figure 2.3: Difference between Internalists and Externalists .................................... 49 

Figure 2.4: Difference between PT’s Top-down Bottom-up Approach ...................... 52 

Figure 2.5: Locating the subject on Perception Theory's Quadrants ........................ 53 

Figure 3.1: Map of Research Methodology .............................................................. 59 

Figure 4.1: Top 4 benefits of AI within CS operations ............................................... 82 

Figure 4.2: Benefits of AI in CS operations ............................................................... 84 

Figure 4.3: Summary of Participants Risks in AI ...................................................... 86 

Figure 4.4 CS Risks Across TOE and Protection Channels ..................................... 89 

Figure 4.5: List of 14 Risk Mitigation Tactics within CS Operations .......................... 90 

Figure 4.6: Participants outlook on AI improving the banks security posture ............ 95 

Figure 4.7: AI Automation Scale ............................................................................... 98 

Figure 4.8: Participants Response to AI's Impact on Engineers in CS Operations ... 99 

Figure 4.9: Participants Results of AI's Decision-Making Visibility .......................... 104 

Figure 4.10: Summary themes of participants feedback to AI Protection Measures 111 

 

 



 10

List of Tables  

 

Table 2.1: Types of Cyber-Attacks (non-exhaustive) ................................................ 32 

Table 3.2: CS participants profiles ............................................................................ 69 

Table 4-1: Summary of Participants Background and Experience ............................ 77 

Table 4.2: Interview Themes and Rationale ............................................................. 79 

Table 4.3: AI tools and its Uses in CS teams ............................................................ 80 

Table 4.4: Description of Risks to AI and its systems ............................................... 94 

  



 11

List of Abbreviations  

 

AAI – Applied Artificial Intelligence 

AI – Artificial Intelligence  

APT - Advanced Persistent Threats 

BEC - Business Email Compromise 

CPU - Central Processing Unit 

CS – Cyber Security 

DoS - Denial of Service 

DDoS - Distributed Denial of Service 

DL – Deep Learning 

GAI – Generative Artificial Intelligence  

GenAI – Generative Artificial Intelligence  

GDP – Gross Domestic Product 

HR – Human Resource  

IIF- Institute of International Finance 

IT – Information Technology  

KPI – Key Performance Indicator  

MA – Metaheuristic Algorithms  

MitM - Man-in-the-Middle 

ML – Machine Learning 

NAS - Network Analysis Systems 

NPS - Network Protection Systems 

PAM – Privileged Access Management 



 12

POPIA - Protection of Personal Information Act 

SaaS - Software as a Service 

SABRIC - South African Banking Risk Centre 

SA – South Africa  

TAM - Technology Acceptance Model 

TOE - Technological Organisational and Environmental Framework  

XAI - Explainable AI 

XXS - Cross-site Scripting 

 

  



 13

Chapter 1: Introduction  

1.1 Statement of Purpose  

The financial services industry in South Africa (SA) is at the helm of the digital 

revolution. The banking sector prioritises digital transformation at the heart of its 

strategy within the financial services sector. In SA alone, the Top 4 banks, in order of 

asset holding size, Standard Bank, First Rand, ABSA and Nedbank, account for over 

50 percent of the continent's assets as Tier 1 banks. They have a combined user base 

of 20 million and contribute around 88 percent to the country’s Gross Domestic Product 

(GDP) (Everington, 2024). The ease of internet and mobile banking across all 

platforms and devices has made transferring capital and data easy and efficient. 

However, the underside of digital transformation is a massive surge in cybercrime, 

resulting in financial losses year-on-year. SA’s banking sector is one of the primary 

targets of cyberattacks. This was witnessed by SA having the third most cyber-crimes 

globally since 2020 (Bank, 2022). The rise in cybercrime warrants an increase in 

cyber-defence and cybersecurity (CS).  

 

Four main technology trends are employed within cyber-security: Cloud and edge 

computing, Applied Artificial Intelligence (AAI), Next-generation software development 

and Trust architecture and digital identity (Atkins L. B., 2024). Of the four, AAI is most 

relied upon in the South African context and will form the focal point of this paper.  As 

such, this paper seeks to ascertain the concatenation of AAI and cybersecurity and 

whether the former acts as a solution to the latter, an enabler to cyber-crimes or if it 

has a benign position but is adapted and implemented due to seeming populus 

consensus and the necessity of being on par or ahead of the digital curve and ahead 

of competitors. This will be achieved via a bi-pronged process. 



 14

1.2 Background of the Study  

The deployment and use of AI in cybersecurity developed alongside and in conjunction 

with key AI disruptors in the banking sector. Today, AI is a part of the DNA of 

cybersecurity itself to the point of almost unconscious relegation. Innovation and 

strategic intent promote the interplay of critical cognitive functioning to support and 

advance machine learning. This paper seeks to question the efficacy of the impact of 

AI in CS. Are AI, AAI, and Generative AI (GAI) trends, necessities, or impediments in 

CS banking operations today? The impetus for the questioning lay within personal 

experience in the banking sector and personally witnessing the radical adoption of AI 

without testing its impact against higher intent, that is, safeguarding and protecting 

data and minimising threats. To ascertain the efficacy, the vantage point is unpacking 

the evolution of AI itself. How did AI become the ethos of the banking sector in SA, 

how did cybersecurity develop to the digital boom, and what challenges AI presents to 

CS operations? These questions will form the basis of the background and precede 

the core and supplementary research questions that follow.  

 

AI has been one of the biggest disruptors in the banking sector. It has radically evolved 

from work automation and data analysis to machine learning, helping to create a 

unique personalised customer experience, to risk management and fraud prevention 

(Atkins L. et al, 2014). “The emergence of AI is disrupting the physics of the industry, 

weakening the bonds that have held together the components of the traditional 

financial institutions and opening the door to more innovations and new operating 

models” (Deloitte, 2025, p. 1). There have been four marked disruptions in the banking 

sector due to AI. Understanding the four points of disruption will assist in identifying 

the needs and importance of cybersecurity within banks.  



 15

Firstly, the impact of Big Data changed the experience and expectations of customers 

as an increasing number of people shifted towards the use of the internet and mobile 

devices. Big data analytics was able to collate unstructured data across multiple user 

platforms systematically and offer highly personalised experiences viz-a-viz traditional 

blanket offerings (Deloitte, 2025). Banks now adopt a 360-degree view of customers' 

interactions with brands, social media and the interfacing of personal data to inform 

their decision-making processes. The second disruptor came from the availability of 

fast computers, hardware, software and cloud infrastructure (Deloitte, 2025). Banks 

could quickly compute, process and store large amounts of data and Software as a 

Service (SaaS) became a core part of all infrastructure. Third, AI was able to disrupt 

the banking sector by assisting in regulatory obligations. Data collection, storage and 

processing aided in banks being able to quickly meet regulatory requirements, 

radically improving the efficiency and output of back-office functions (Deloitte, 2025). 

The last and fourth disruptor appeared via inter-bank competition. Banks leverage AI 

and other tools to customise solutions via harvesting large amounts of data they 

possess. As a result, there is a race amongst banks to adopt and implementing the 

last tech innovations to create solutions that have a stand-out factor and provided a 

value-added service internally and externally (Deloitte, 2025). Parallel to these 

disruptors was the cultivation of talent management and hiring of Information 

Technology (IT) professional which also gave rise to the gig-economy. Emerging 

technologies offer significant benefits but simultaneously introduce new cyber threats 

and entrench existing ones.   

 

CS risk management and operations are integral to a bank’s core infrastructure and 

support. The importance of a comprehensive strategy has never been more prudent 



 16

as companies aim to increase their digital and technological footprint. With data being 

the new currency, banks are not only combating highly trained, well-funded cyber 

criminals but cyber terrorists as well. Cyber criminals are also adopting emerging 

technologies to support their attacks. According to the CrowdStrike (2024) global 

threat report, “Electronic Crime (eCrime) continues to rise and is the most pervasive 

threat in 2023. Data-theft extortion also continues to rise, and 2023 saw a 76 percent 

increase in victims named on eCrime dedicated leak sites compared with 2022. As 

companies increase their use of technology, they are also increasing the number of 

avenues for a potential cyberattack by mature threat actors.” As banks increase their 

use of digital technologies, they also expand the scope for cyberattacks, especially by 

mature threat actors. Due to the slow implementation of legislation and regulations 

within the AI space, combined with the lack of morality and ethics of AI itself, it allows 

the layman to utilise AI platforms to engage in criminal activity. Anyone with a phone 

can use AI platforms to try to hack systems. According to the 2023 Future of 

Cybersecurity Survey (2024), completed in collaboration between McKinsey and the 

Institute of International Finance (IIF), wherein they interviewed banks globally, the 

biggest risks presented included cyberattacks, AI, talent management, third-party and 

supply chain management, and data security as seen in Figure 1:  

 



 17

 
Figure 1.1:Top three cyber risks in the financial services sector 

 

The dynamic interconnectedness of the threats is tied to the four digital disruptors 

mentioned earlier. What risk does big data and cloud storage present to banks' CS 

operational support? While banks adopt the latest talent and technological capability 

in creating complex firewalls to prevent breaches, banks are also reliant on third-party 

data, and any weaknesses in their system can result in data breaches and data theft. 

This is corroborated by the above-mentioned survey, wherein most survey 

respondents “Recognised the need to strengthen critical cybersecurity capabilities, 

including third-party or supply chain management and privileged access management 

(PAM).” The importance of cloud storage cannot be underscored, given that data is 

the era's currency. Any disruption of files, poisoning of data, and leakages could have 

massive legal ramifications for banks and financial losses for customers. SA has seen 

a persistent annual increase in cybersecurity attacks since 2019 (Mugwagwa, 2024) 

According to the South African Banking Risk Centre (SABRIC), South Africa lost more 

than 3 billion Rand in 2023 due to financial crime. Reports from the South African 

Cyber Security Hub and cybersecurity companies indicate that attacks targeting 



 18

individuals, businesses and government departments are on the increase (SABRIC, 

2023). 

 

Before banks introduce newer and newer technologies, there needs to be measures 

to ensure the risk management of a capability is exercised to the superlative, the 

ramifications of which, when not performed, result in the risks outweighing the benefits.  

Additional risks include corporate espionage and headhunting/talent management, 

which can easily lead to some banks monopolising the market. Threats may not only 

come from external sources but also internally and from other banks. It can also result 

in the prevention of centralising key functions and limit cross-pollination for the 

advancement of the industry. This also puts pressure on internal teams to always 

adopt the latest tech instead of creating vertical growth. Lastly, the speed at which 

regulators create new regulations to adapt to changes in the industry is far slower than 

the fintech evolution. Resultantly, leveraging AI will always result in banks being further 

ahead than regulators or managing ways to outmanoeuvre regulators. Banks are 

willing to pay the fines regulators impose because the outcome and financial reward 

far outweigh the cost of fines. This protects banks, gives more liberty to AI and causes 

regulators to lose credibility. Banks also hire the best talent in AI, and regulators 

struggle to secure the personnel needed to effectuate regulations promptly.  

 

While CS is a critical part of business operations, as technology evolves and the ethical 

concerns around AI are magnified, CS operation officers are questioning the use of AI 

and GAI in preventing cyber-attacks. Naturally, without any CS support, the banks’ 

entire structure and well-being will collapse, so the expulsion of AI in CS is not an 

option. However, it is becoming increasingly visible that AI as a solution is not 



 19

preventative and only superficial. Furthermore, CS teams are underfunded and face 

challenges when requesting additional funds to set up the infrastructure to limit cyber-

criminal activity. Acknowledgement of underspending in capabilities has grown. 

According to the Future of Cyber Security Survey (2024), 70 per cent of the survey 

respondents believe they are underspending and should spend more. The marked 

shift between the 58 per cent who acknowledged underspending in 2020 and the 70 

per cent in 2023 alludes to companies suffering the consequences of underspending, 

as depicted below.  

 

 
Figure 1.2: Percentage of Fin companies underinvesting on CS capabilities 

 
 

Financial services institutions allocate approximately 13 percent of their IT budget to 

CS.  

 

Banks are not conducting exercises to ascertain how much capital will be lost with 

weak, fair and strong CS protection and ratioing it to the capital gains incurred. The 



 20

rapid pace of technological evolution is stripping time away from teams to focus on 

quantitative metrics to assess the impact – financial, ethical; operational, talent 

retention and upskilling and so forth – of AI of CS functionality and operationality. As 

banks continue investing heavily in technologies, there is now a warranted need to 

consider the short, medium and long-term implications of these technologies for CS to 

maintain protection of their environments and to question novel capabilities outside of 

AI that can prevent cyber security theft and enable improved protection of data and 

resources (Atkins L. B., 2024).  

 

Thus, financial institutions face various cyber threats that are both sophisticated and 

diverse. The volume of attacks and the increased complexity pose a significant risk. 

The increased number of successful cybersecurity attacks has led to increased 

sensitive data, critical infrastructure and financial asset compromises (Kayode-Ajala, 

2024). Perpetrators of these attacks range from individual criminals to state-sponsored 

hackers. Threat actors employ a wide range of methods to attack financial institutions. 

Attacks can vary from basic phishing attacks to more sophisticated ransomware 

campaigns. As banks are using AI to innovate and increase their technological 

footprint, profitability and scalability, it is equally being adopted by cyber criminals to 

leverage key data points and attack resources and capabilities. These actions form 

causative questions: Is the AI used by the banks CS operational teams effectively 

combating the root cause of criminal cyber-attacks? Is a technological AI trimming 

needed to improve operations? What non-AI solutions and capabilities can be adopted 

to improve CS? This thinking form will formulate the basis of the research problem, 

questions, and the following justification.  

 



 21

1.3 Research Problem 

The use of AI has the potential to solve and address many of the cybersecurity 

concerns and challenges which the banking sector in South Africa is currently facing. 

However, AI, like any other technology, is also vulnerable to exploitation and 

cyberattacks (De Azambuja, 2023). Knowing the new threats and risks introduced 

when AI-powered technologies are used remains paramount.  

 

An analysis of the literature indicates that there is a research gap regarding the use 

and impact of AI in cybersecurity and South African Banks. Therefore, it is necessary 

to investigate the impact of AI-powered cybersecurity solutions on the number and 

severity of cyber threats in South African Banks year-on-year. 

 

The purpose of this research is to explore the impact AI has on improving the 

operational efficiency of cybersecurity in banks. While the adoption and 

implementation of AI across banks is rampant, does AI actually decrease the CS 

team’s workload and, based on machine-based learning and predictability, engage in 

smart learning to prevent attacks, or is it merely a tool/resource for responding to 

threats? The uncertainty and lack of clarity on the highly pertinent and impactful matter 

necessitated the investigation. 

 

1.4 Research Questions  

 To what extent can artificial intelligence serve as an effective solution for 

eradicating and preventing cybercrime within the South African banking sector? 

 What are the factors that contribute to the decrease of the CS staff complement 

in banks to improve ‘organisational effectiveness’? 



 22

 What evaluation criteria or assessment frameworks are employed by banks to 

assess the effectiveness of AI-CS capabilities in reducing cybercrime through 

preventative measures? 

 

1.5 Justification for the study  

The impetus and justification for the study are based upon personal experience in 

cybersecurity in the banking sector in SA. While AI offers radical approaches to 

computing, processing and storage, it also offers innumerable challenges, lacks 

ethical and moral considerations and legislation, and provides little to no ramifications 

to cyber criminals during data theft and breaches. CS teams are often left in isolation 

when breaches occur due to AI. The justification is predicated upon understanding 

both ends of the AI coin.  There are three main reasons why this topic has been chosen 

to explore.  

 

Firstly, AI is seen as a turnkey solution; that is, implementing AI capabilities in CS will 

eliminate any challenges and issues. AI does possess several advantages, especially 

regarding process, synthesising and collating email, endpoint, mobile, network 

perimeter security, for example, Distributed Denial of Service (DDoS), but most 

importantly, it does not eradicate threats from happening, which is the main function 

of CS. Secondly, machine learning is causing job losses, with only 50 per cent of a 

team needed compared to what was. Given the thin regulation around ethics in AI, 

with a smaller CS staff complement, data quality, protection and sanctity are highly 

comprisable compared to an in-person staff complement that can exercise ethical 

considerations and make decisions in line with business outcomes. From a human 

capital perspective, the impact on people needs to be considered in the medium-to-



 23

long term, and not solely in the short term, as most people/organisations are currently 

doing.  

 

Third, Security AI systems are SaaS, which are sitting in the Cloud. This begs the 

question: Are banks fully aware of the telemetry that the AI system is collecting, where 

it is stored and who has access to it? Do they understand the potential impact if the 

data collected by these AI platforms land in the wrong hands? An ongoing supply chain 

and third-party risk need to be considered. Using highly specialised capabilities for 

criminal activity will always be open and an option. AI can be used to compromise 

security AI systems and to poison data used by AI algorithms. AI, like any other 

technology is vulnerable to misuse and attacks. 

 

The adoption of AI in CS is a trend. But once an organisation employs AI to a certain 

degree, there is a point of no return. This paper adopts the view that AI CS platforms 

do not improve a bank's cybersecurity posture as popular consensus dictates, but AI 

becomes the oxygen for digital firms to a point of cessation. Today, big banks have no 

choice but to adopt AI technologies. With an already laid-out trajectory, will AI/CS 

operate on a loop-based continuum ad infinitum?  

 

1.6 Delimitations of the Study 

The key limitation of the study emerged in the disparity between access to private 

quantitative data within banks' CS teams and the qualitative methodological approach 

adopted to ascertain results. The primary intention of this paper is to ascertain if AI 

has a real impact in preventing cyber-crimes, saving organisations money and 

developing critical infrastructure to limit the attacks against banks due to impenetrable 



 24

protections. Unfortunately, access to this data amongst all big banks is highly 

confidential and public knowledge or it will jeopardise their overall CS. As a mitigatory 

response and methodological substitute, this paper employed interview questions with 

key personnel in CS in the banking sector to gauge their sentiment on efficacy, 

impacts, and solutions.    

 

What follows is the Literature Review, which explores a holistic view of the interplay 

between AI and CS in a macro- and micro-economic landscape.  

 

1.7 Report Outline  

This paper will adopt the following chronological structure to answer the research 

questions.  

 

Chapter 1: Introduction 

This chapter contextualises the research questions and provides a background and 

landscape to the operating dynamic of cybersecurity professionals. It laid the 

foundation for the rationale for the research topic and positioned the research 

questions and their concomitant justifications. 

 

Chapter 2: Literature Review  

The Literature Review offers both a topographical and in-depth view of the body of 

literature that exists on the subject matter. It will specifically look at the evolution of AI 

adoption within SA banks with the key tools, processes and people that change along 

the way. The literature is categorised and analysed via the Technological 

Organisational and Environmental (TOE) framework as well as a psychological-



 25

qualitative approach called Perception Theory. The usage of Perception Theory is 

novel to this paper and the industry as traditionally only technologically oriented 

frameworks have been used for technological analysis.  

 

Chapter 3: Research Methodology 

This paper has adopted a qualitative phenomenological approach to analysis. Twelve 

participants were chosen as the sample group, and they occupy diverse and various 

roles in CS operations within SA banks. Via semi-structured interviews, the 

participant's experiences and tenure in AI-CS provide an insightful, rich and accessible 

view on the intricacies and sensitivity of AI-CS operations.  

 

Chapter 4: Findings of the Study  

Chapter 4 offers an in-depth view of the answers provided in the interviews. The 

participants were asked questions that broadly but succinctly covered the internal 

operations of a CS team within a big bank. The data was categorised into themes, and 

the views of TOE and Perception Theory were used to analyse the findings. The 

questions were all curated to cover the research questions and very interesting 

outlooks that both confirmed and challenged the narrative were discovered.  

 

Chapter 5: Conclusions and Recommendations  

The paper concludes with an overview of how the research questions were answered. 

It presents a sobering view on the reality of AI-CS operations and mentions gaps within 

academia and the technical CS teams that future research can explore. Lastly, a 

framework was created that showed how data travels across CS operations and the 

risks and benefits that teams and organisations face in the data journey.  



 26

Chapter 2: Literature Review  

2.1 Introduction 

This Literature Review aims to review the existing body of work on the subject matter 

delineated. While there is an exceptionally large body of academic and technical 

literature on AI itself across all sectors globally, the interplay of AI in cybersecurity 

operations is relatively niche. Furthermore, the determination in assessing the 

‘perceived impact’ of AI in CS operations within banks was remarkably thin and 

infantile. This presented two core challenges. The first challenge was the ultra-

technical nature of AI CS data. While the body of literature was thin, the journal articles 

and online sources offered a very in-depth view of the software being used, but it solely 

mentioned the solutions being used and the rationale – a digital evolutionary mapping. 

This information was vetted by the researcher, who has direct experience in bank CS 

operations. The TOE framework will be used for its holistic outlook in structuring the 

ideology of AI adoption in CS operations. The framework also assists in answering 

RQ2.  

 

The second challenge, which was more acute, was the subjective nature of assessing 

the impact and adoption of AI in CS within banks in SA. There was a lack of succinct 

theories and frameworks to draw upon that assisted in answering the research 

questions. The Technology Acceptance Model (TAM) was initially explored for its 

acceptance usage framework. Still, it did not suffice to guide the core research 

questions of whether or not AI is supporting CS operations. Looking outside of 

technological approaches and frameworks, this paper adopted Perception Theory as 

the guiding framework in gauging the efficacy of AI solutions in CS operations. 



 27

Perception Theory, with its qualitative phenomenological outlook, has dramatically 

guided the narrative and opened the channel to inter-disciplinary pollination and 

provided a unique philosophical-psychological outlook completely unexplored in digital 

and AI approaches and frameworks. This has also informed the framework-come-

theory postulated and positioned in Chapter 5, respectively (RQ3).  

 

The structure of the Literature Review is head-first to explore the vicissitudes and 

advancements of AI in cyber-security, followed by a critical analysis of Perception 

Theory. The Literature Review will conclude with a synthesis of AI-CS and Perception 

Theory.  

 

Before continuing, it is important to mention that the taxonomy of AI can be dissected 

across themes and applied to all sectors of the economy, from military advancements, 

determination of cases in the healthcare sector and so forth. The AI lexicon is vast. 

The term AI encompasses smartphone usage to metaheuristics, and a definition of the 

term itself is evasive despite standardisation attempts by governments, industry 

leaders and digital organisations. Inspired by Kelly, Kaye and Oviedo-Trespalacios 

(2023), the following definition of AI will be used in this paper: “An unnatural object or 

entity that possesses the ability and capacity to meet or exceed the requirements of 

the task it is assigned when considering cultural and demographic circumstances.” 

(Kelly, 2023, p. 2). There is an exceptionally large body of literature available on AI 

itself, should the reader wish to explore, but given the vastness of AI, for the purposes 

of this paper, AI will be contextualised in the cybersecurity operations within South 

African banks and will not be explored any further.  



 28

2.2 Defining the TOE Framework 

TOE examines the adoption of technology from the viewpoint of an organisation. It 

outlines the various aspects that impact the adoption of technology and the probability 

of its implementation. The theory suggests that the technology adoption processes of 

an organisation are influenced by its technological, organisational and environmental 

settings (Hasani, 2023). 

 

The technological context encompasses the internal and external technology pertinent 

to the company. Technologies encompass both physical equipment and procedural 

methods. The organisational context refers to the attributes and assets of the 

company, such as its dimensions, level of centralisation, level of formalisation, 

managerial framework, workforce, quantity of surplus resources, and connections 

among employees. The environmental context encompasses the dimensions and 

composition of the industry, the firm's rivals, the overall economic conditions, and the 

governing framework. The TOE framework offers a comprehensive viewpoint on 

technology adoption, adaptability and application (Baker, 2011). 

 

The TOE framework is relevant to this study from a technology perspective because 

AI, as a technology, can help transform CS operations and help assess the readiness 

and compatibility of AI-CS solutions within local banks in SA. From an organisational 

level, the TOE framework evaluates the factors that influence the adoption and impact 

of AI on banks, such as resource constraints, skills shortages and legacy systems, to 

name a few. South African banks are highly regulated and operate in an environment 

with a growing need for financial inclusion. From an environmental perspective, the 

TOE framework considers external pressures such as regulators, legislations and 



 29

policies, customer expectations and inter-bank competitiveness. All three features will 

be discussed chronologically.  

 

2.3 Technological component of TOE affecting CS operations  

AI has the potential to transform cybersecurity and improve its efficiency and 

effectiveness. AI-based security software enables faster detection and response to CS 

threats. It automates repetitive tasks, freeing security professionals to concentrate on 

more intricate security-related tasks (Sontan, 2024). AI systems can filter through 

massive amounts of data and identify patterns that point to possible threats. The 

adoption and use of AI-powered systems in CS is expected to significantly increase 

globally. Markets and Markets predicts that the global market for AI in cybersecurity 

will increase from $8.8 billion in 2019 to $38.2 billion in 2026 (a 23.3 per cent increase) 

(Ibrahim, 2023). Organisations use AI to improve and strengthen their cyber defence 

systems against cyberattacks. AI benefits businesses that use it in their cyber defence 

operations. The advantages include automation, threat intelligence, and enhanced 

CS. AI-based CS tools are increasingly important for identifying and averting 

cyberattacks (Ibrahim, 2023). 

 

The growing use of online services has seen an increase in the frequency and 

sophistication of cyberattacks, resulting in an obsolescence of traditional CS solutions 

(Kaur, 2023). Networks and system complexity are increasing, and AI-powered 

cybersecurity solutions have become essential. The adoption and use of AI tools 

provide considerable advantages to organisations in assisting them in protecting their 

networks from cyberattacks (Jada, 2024). 



 30

2.3.1 The necessity of Cybersecurity  

In today’s rapidly advancing digital age, the CS landscape confronts unprecedented 

cyber threats. Traditional security measures, such as firewalls and intrusion detection 

systems, have fallen short in their advancement as they require manual processes 

and systems drastically slower than the automation and speed of data processing 

offered by AI software and solutions. The COVID-19 pandemic exacerbated the 

adoption and scaling of AI as businesses of all sizes shifted to an online presence and 

banks experienced a surge in activity to cater to customer needs and improve 

competitiveness. Simultaneously, this led to a surge in cybercrimes, which is predicted 

to reach $10.5 trillion in 2025 (Eian, 2020). To mitigate these losses, there is a poignant 

drive to improve AI-related resources and tools to improve banks' cyber defence. AI 

mimics and simulates human behaviour, resulting in the speed of automation beyond 

manual human capability and with profound benefits (Xu, 2021). The evolution of AI 

from Machine-Learning (ML) to Deep-Learning (DL) has minimised human error and 

effort. It has become an invaluable tool to the myriad forms of cyber-attacks banks 

face daily.  

 

How society functions has also rapidly changed. Customers would traditionally only 

go to a brick-and-mortar bank or telephonically discuss certain mandated matters, but 

today, all forms of banking can be performed online. Resultantly, cyber-attacks disrupt 

and target computer systems, networks and data, and stringent security measures 

exist along these triadic platforms (Salem, 2024). Tabulated below is a non-exhaustive 

list of cyberattacks: 

 

 



 31

Type of Attack Description 

Phishing  Tricking individuals into divulging sensitive information such as login 

credentials and financial data by masquerading as trustworthy electronic 

communication (Forbes, 2024). 

Malware Malware/ ‘malicious software’ includes viruses, worms, trojans, and 

ransomware that are designed to damage or unauthorised access to 

computer systems and data (Forbes, 2024). 

DDoS Denial of Service (DoS)/Distributed Denial of Service (DDoS) attacks 

overwhelm a system's resources, making it unable to respond to 

legitimate service requests. DDoS attacks use multiple compromised 

computer systems as sources of attacks, exacerbating the scale and 

impact of assault (Forbes, 2024). 

MitM Man-in-the-Middle (Mitm) attacks that are invisible, intercept 

communication between two parties, resulting in theft, alterations or 

fabrication of messages, leading to data breaches and eavesdropping 

(Forbes, 2024). 

SQL Injection Exploits vulnerabilities in a database-driven website by injecting 

malicious SQL statements, resulting in access to sensitive data that can 

be tampered with and manipulated (Forbes, 2024). 

Zero-Day Exploit On the ‘zeroth’ day, a vendor becomes aware of a vulnerability. The 

attack exploits vulnerabilities before they can be addressed, 

exacerbating the ability to defend (Forbes, 2024). 

Ransomware A form of malware that encrypts a victim's files, making them inaccessible 

and holding the victim hostage until they pay for the decryption key 

(Forbes, 2024). 

XSS Cross-site Scripting (XSS) attacks third-party web resources to run 

scripts in the victim's web browser or app (Forbes, 2024). 



 32

APT Advanced Persistent Threats (APT) occur when an individual or group 

acquires unauthorised access to a network for an unnoticed period of 

time, resulting in the exportation of data (Forbes, 2024). 

BEC Business Email Compromise (BEC) attacks target employees with 

financial authority to trick them into sending money into the attacker's 

account (Forbes, 2024).  

Crypto-jacking  Confidentially uses a victim's computing resources to mine 

cryptocurrency. 

Password 

Attacks 

Attempts to crack a user’s password and phishing for passwords. 

Hacking tools include Brute-Force, Rainbow Table, Credential Stuffing, 

Password Spraying and Keylogger attacks (Forbes, 2024). 

Insider Threat Individuals within an organisation  who misuse the company’s systems 

and data. 

Botnet Attack A network of compromised computers controlled remotely by an attacker 

to coordinate malicious attacks (Forbes, 2024). 

Table 2.1: Types of Cyber-Attacks (non-exhaustive) 

 

“Intending to cause damage, unauthorized access, or service interruptions that cause 

severe data loss or financial damage and often lead to long-lasting consequences, 

these are the insider threats that represent a significant and growing segment of these 

attacks, usually committed by disgruntled or rogue employees who exploit their 

authorized access to steal data or cause harm. These threats can also emerge from 

intrusive applications that users accidentally install on their devices, allowing these 

apps to access and misuse sensitive information. Advanced behavioural anomaly 

detection and auto-resiliency mechanisms are being developed to combat these 



 33

threats by proactively identifying and mitigating malicious actions at both the employee 

and application levels.” (Salem, et al, 2024, p.4). 

 

Irrespective of the nature of the attacks, AI can learn from data, compute, process and 

defend against threats and attacks, enabling the protection of infrastructure, software 

and data. The benefits of integrating AI with CS include improved decision-making due 

to concise quantitative data being easily accessible, enhanced detection of network 

intrusions, overall management of cyberattacks, and its impact on systems and 

software. This evolution, which allows for real-time detection and response, has 

significantly reduced the ‘false positives’ rate, which is common in traditional models 

of CS. ML and DL have enabled predictive analysis to be made of datasets and to 

identify potential vulnerabilities before they are exploited. This has enabled a 

preventative, proactive approach, as opposed to a reactive one (Welukar, 2021). For 

example, phishing attacks have become increasingly sophisticated, and traditional 

methods of detection have been unable to identify and prevent them effectively. 

Conventional methods relied on blacklists, heuristics, and content analysis but 

attackers are now using AI and social engineering resulting in convincing phishing 

emails that easily evade traditional detection methods (Eze C. S., 2024).  

 

The three main types of AI in CS that will be explored are ML, DL and Metaheuristic 

Algorithms (MA). 

  

2.3.2 AI CS Anthropomorphism  

AI integrates vast quantities of data with fast, iterative processing and sophisticated 

algorithms. This allows the software to acquire knowledge automatically by analysing 



 34

patterns or characteristics in the data. AI is a discipline within computer science that 

spans several domains, such as ML, DL, data analytics, languages, and software 

engineering (Shrivastava, 2024). These fields frequently entail the creation of AI 

algorithms that mimic the decision-making mechanisms of the human brain. These 

algorithms can learn and remember information from available data, resulting in more 

accurate categorisation (Shaveta, 2023). 

 

Vulnerability management and remediation of vulnerabilities in operating systems and 

applications have become increasingly difficult, with organisations struggling to 

manage the ever-increasing number of vulnerabilities due to traditional models 

(Kumar, 2023). AI assists banks in better managing and remediating vulnerabilities 

across their networks. AI-powered vulnerability management systems can scan 

networks proactively for potential vulnerabilities and can detect and classify network 

vulnerabilities (Rizvi, 2023). AI-driven incident response platforms utilise sophisticated 

algorithms, such as ML and natural language processing, to examine security alerts, 

rank incidents based on importance and carry out predetermined reaction steps 

independently. These platforms can process high numbers of real-time security alerts 

and make data-driven decisions to promptly and efficiently resolve security events 

(Sontan, 2024). Using AI-powered incident response systems reduces human error, 

and the speed at which incidents are responded to is improved. The organisation's 

overall security posture is improved as AI-powered incident response systems are 

more flexible and scalable. AI-powered incident response systems can also assist with 

anticipating and preventing future attacks and improve situational awareness and 

threat visibility (Rizvi, 2023). The first shift from traditional AI-related vulnerability 

management and remediation was to Machine Learning. 



 35

2.3.2.1 Machine learning 

ML empowers computers to solve and interpret issues without extensive 

programming. The learning methodology analyses past data, consolidates various 

algorithms, identifies them and demarcates them into categories/typologies, and 

performs tasks based on the desired outcome (Kuntla, 2021). The algorithms are tri-

pronged (see Figure 2.1 below), and cluster data to defend against the attacks 

mentioned in Table 2.1 above:  

 

 
Figure 2.1: Types of Machine Learning 

 

ML architectures are broadly scoped and focus on identifying false data to censor 

outcome manipulation. They detect network anomalies and internally function like an 

immune system by defending against alien or foreign data and software.  

 

While ML was promising for its duration as the forerunner of AI in CS, its reliance on 

manual feature extraction lay too close to traditional modelling, which limited the 



 36

efficiency and accuracy of threat detection and became a limitation within CS. ML’s 

functionality is significantly dependent on the precision of feature extraction and 

recognition (Eswaran, 2023). This is extremely important as biometrics has become 

the access format of customers within banks.  

 

2.3.2.2 Deep Learning 

Disruptions and growth of technologies within AI are almost always a result of shifts in 

the global/macro-operating environment. Akin to CS responding to threats in a reactive 

way, the development of AI has produced novel solutions reactively. The IoT, the 

explosion of network connectivity, positioned CS at the forefront of operations due to 

the large amounts of data that needed to be stored on the cloud. Malware and third-

party cloud computing and data storage suppliers became the new target for cyber-

attackers. Attackers would look at networks and malware points of vulnerability, and 

where ML had reached its zenith in protectionism. Herewith, DL presented an 

anthropomorphic response to the challenges, CS methodologies, and operations as a 

whole (Morovat, 2020).  

 

DL grew from some of the challenges of ML. It revolutionised the AI landscape by 

creating “Advanced neural networks that can process large amounts of data and learn 

from experience, mimicking human brain functions to recognize complex patterns” 

(Barik et al, 2022). It mimicked images, voices and behaviours and open pathways for 

changes in robotics, speech and facial recognition. In the sphere of CS, DL plays a 

critical role in detecting intrusions and monitoring for malware (Musa, 2024). It was 

propelled by advancements in preventing APT attacks by recognising different scales 

of attacks used in the most evasive tactics (Salem et al, 2024). As presented in Figure 



 37

2.2 below, the neural network has several layers, all highly complex and hidden in 

nature. These barriers prevent the breach of attacks, help systematise and categorise 

the attack type and configure and learn based on the inputs to adapt to the rampant 

and diverse types of attacks across all three malware platforms, network and data.  

 

 
Figure 2.1: Difference between Simple and Deep Learning Neural Networks 

 

Naturally, given that customers are providing access to their biometrics, there are 

stringent requirements for data protection, as loss of data could result in a customer’s 

breach of privacy and hackers and cyber thieves using the data to gain access to 

customers personal information resulting in identity theft, citizenship mimicking, bank 

account information and social media accesses. A significant and tight regulation is 

needed on this account. The necessity of CS was never more prudent. As the bank's 

CS teams improve their solutions and offerings, cyber thieves also use the latest AI to 

engage in criminal activity, creating a hostile environment, but also being the impetus 

for ongoing progress and development. The next type of AI comes in the form of MA.  

 

2.3.2.3   Metaheuristic Algorithms 

MA addresses complicated problems that cannot be solved using standard processes.  



 38

MA are optimisation strategies inspired by animals' problem-solving activities and 

abilities in their natural habitats. The algorithms have shown remarkable optimisation-

related success when applied. MA is delineated into four categories, namely evolution-

based, swarm intelligence-based, physics-based, and human-related algorithms as 

shown below: 

 
Figure 2.2: Four Types of Metaheuristic Algorithms in CS operations  

 

MA is vital in improving the efficiency and accuracy of various detection learning by 

enhancing the learning as they expand the search space explored during model 

training, potentially uncovering superior solutions that traditional methods might miss. 

This is particularly beneficial in CS, where the landscape and attack patterns change 

rapidly. The advantages of MA in cyber-attack detection and defence include, 

according to Nordin et al (2021) in A comparative analysis of metaheuristic algorithms 

in fuzzy modelling for phishing attack detection, an improvement in optimisation, 

automation and speed. 

 

 Optimisation: MA’s are better at finding optimal solutions to complex problems 

that are otherwise too challenging for conventional methods;  



 39

 Automation: By automating the tuning of detection parameters, the algorithms 

minimise the need for human intervention, making the detection process fast 

and reliable;  

 Speed: MA achieves faster convergence to effective solutions, essential in time-

sensitive cybersecurity environments where threats must be quickly identified 

and mitigated. 

 

Collectively, these AI-based technologies address the challenges in CS more 

effectively than traditional methods. They do so by automating the detection and 

response processes and enhancing the speed and accuracy of threat detection. 

Understanding the mechanisms and operations behind these AI models offers more 

profound insights into their application, highlighting their critical role in developing 

more resilient cybersecurity systems. MA supports the robustness and adaptability of 

cyber-attack detection systems, making them more effective against various cyber 

threats. They ensure that detection systems are accurate and remain relevant as 

attack methods evolve. 

 

Interestingly, the results of AI were experienced similarly in Qatar's banking sector. In 

a study conducted by Al-Dosari et al (2022) it was revealed that AI is a crucial 

technology for enhancing the cybersecurity measures and posture of banks in Qatar. 

In another study done by Ibrahim and Hassan (2023) it was found that established 

businesses were using AI to boost their defences against new emerging threats. AI-

powered technologies is rapidly changing cybersecurity, making it more effective and 

efficient. AI can utilize predictive modelling to proactively prevent threats, identify 

threats in advance and proactively prevent them by analysing previous attacks and by 



 40

identifying patterns. AI can quickly assess security risks and act on potential threats in 

real time (Rizvi, 2023). AI is now used for real-time threat detection and prevention, as 

it can continuously monitor network traffic and identify suspicious user behaviour 

through AI-powered behaviour and anomaly detection systems (Sontan, 2024). AI CS 

technologies can help enhance threat detection, vulnerability analysis, and incident 

response.  

 

2.4 Organisational component of TOE affecting CS operations 

Every organisation, including banks, has certain goals and objectives to achieve year-

on-year. In SA alone, the top-tier banks possess a healthy level of competition, which 

promotes the advancement of AI and digital tools and solutions. The banks have 

become the entry point and testing vehicle of anthropomorphic AI deployment. The 

influence of AI cuts across all sections of the bank, from the increase in productivity in 

legal departments due to AI storage, processing and data tools, to brick-and-mortar 

stores increasing the number of digital features in the stores to improve customer 

service. AI has infiltrated all parts of the bank as an ideology and way of operating. In 

this context, four points will be analysed: the interconnectivity of AI, third-party 

vendors, CS security personnel and the gig-economy.  

 

AI as a tool has a scalable feature. Depending on the nature of work to be completed, 

anything from data processing and storage can be performed, up to highly complex 

malware and Central Processing Units (CPUS). The scalability and interconnectivity 

of AI allow for a centralisation functionality that accords CS officers oversight over 

data, network, and malware intra-organisational protection (Ahmed et al., 2022). 

“Network protection using AI-driven network analysis systems (NAS) and network 



 41

protection systems (NPS) can guarantee the safety and availability of computer 

networks within an organisation, not just for a single computer but for an entire 

computer network system simultaneously.” (Xu, 2021). AI solutions can be adopted at 

any juncture within the life cycle, allowing for a complete interconnected solution. Each 

feature exists as a totalised, individual whole, while belonging to a bigger totalised 

centralised system (Raimundo, 2021). The usage of third-party vendors has 

necessitated the need for greater protection. While internal threats can be detected 

and managed, external threats from vendors has the ability of crippling the system. 

CS defence and protection must also extend beyond the parameters of the banks, 

including the vendors.  

 

The use of AI has an impact on the people supporting cybersecurity platforms. Due to 

its rapid growth, AI technology has created gaps by displacing the market’s knowledge 

and understanding of operational procedures. To stay employable and competitive in 

the field of CS, professionals need to learn about AI and gain expertise on AI-powered 

platforms continuously. Having the right people to implement and support AI-powered 

cybersecurity platforms in an organisation is critical. Additionally, and sub-tangentially, 

employee-training programs are essential, focusing on the significance of 

cybersecurity knowledge to deter phishing attempts and other forms of social 

engineering approaches (Andriu, 2023). In a study conducted by AL-Dosari et al. 

(2022) and Ibrahim and Hassan (2023) it was found that there is a lack of skilled 

employees who can deploy, and support AI-powered cybersecurity platforms. The gig-

economy, the notion of a global citizen and the outsourcing of IT specialists were 

common practices between 2010 and 2020; the COVID-19 pandemic magnified the 

practice. While retaining key talent and finding the right people in person may be 



 42

challenging, remote working has allowed banks to rely on personnel in other countries 

for the experience and expertise in managing the systems and processes.  

 

Lastly, there is the looming concern of relevancy and machines replacing humans. 

Time-consuming and manual tasks are now being performed by AI, displacing human 

labour. There is an ongoing debate and thoughts of regulation to prevent the wholesale 

machine-replacing-man narrative, the results of which have massive macro socio-

economic consequences.  

 

2.5 Environmental component of TOE affecting CS operations 

AI collects and processes large volumes of data, and personal data usage in AI models 

is growing, which presents significant challenges to sensitive data protection. The data 

protection used by AI has become increasingly concerning (Sontan, 2024). Protecting 

this data is critical, as misuse or exposure can impact privacy and individuals’ rights. 

There is a mandatory need to protect AI privacy as a fundamental right. Organisations 

must address the associated privacy challenges and take steps to protect personal 

information. AI regulatory and ethical frameworks can guide to ensure the ethical and 

responsible use of AI (Ferrara, 2023). However, on par with international standards, 

there is no regulation on AI. Presently, there are only four Acts that enforce and protect 

customers' data in SA - the Protection of Personal Information Act (POPIA), the 

Copyright Act, the Patents Act, and the Competition Act. 

 

After coming to power in 2019, President Cyril Ramaphosa initiated commissions and 

steps to help regulate AI in the fintech industry, but no formal enforcement has been 

issued to date. Given the absence of standardized regulations for AI usage across 



 43

countries, prospective research can delve into diverse legal frameworks that 

effectively oversee the application of AI in cybersecurity within different nations. These 

studies can potentially mitigate the risk of malicious AI usage in organisations by 

contributing to the establishment of more comprehensive and effective regulatory 

measures.  

 

2.6 Challenges to AI CS operations  

The challenges of AI in CS will follow the same format as TOE.  

 

2.6.1 Technological Challenges  

Human error is the main factor contributing to CS breaches (Amoresano, 2023). 

Humans are considered the most vulnerable point in the cybersecurity system. 

Whether the human error is a decision-based error or an error in performing the task, 

whether that fault is intentional or not, or skill-related, the initial stage in establishing a 

highly secure cyber environment is by eliminating the possibility for human error. 

Automating CS tasks by leveraging AI minimises the requirement for human 

involvement and interaction, reducing potential human mistakes throughout the 

security lifecycle. What sometimes takes hours to do, AI can do effectively and 

efficiently in seconds (Jada, 2024).  

 

The use of AI still faces challenges, as there are harmful activities from telemetry that 

AI does not identify as threats. Adversarial actions targeting ML can manipulate ML 

training samples to influence the ML model’s accuracy (De Azambuja, 2023). These 

attacks aim to disrupt the sample data’s quality, undermining the model’s 

trustworthiness. The goal is to generate malicious behaviour that appears legitimate. 



 44

AI’s potential replacement of security analyst roles, normally responsible for detecting 

and responding to these events, is hindering its effectiveness, leading to uncertainty 

in the risk mitigation strategies of financial institutions like banks. Furthermore, the 

algorithms used by AI tools may raise concerns about their testing, potentially leading 

to bias and some threats not being detected (Ibrahim, 2023). 

 

Vulnerabilities and risks associated with AI platform implementations have grown in 

concern. The manipulation of training data, the initial design of the system, the lack of 

robustness in AI algorithms, and the abuse of implementation weaknesses all lead to 

vulnerabilities in AI (Villegas-Ch, 2023). These risks lead to adversarial attacks, 

unwanted biases, information leaks, and AI model manipulations. Implementing a 

robust security framework is necessary to mitigate these risks (Villegas-Ch, 2023). 

Hackers can exploit AI-based protection systems and evade the AI from detecting 

them. AI-driven attacks use neural fuzzing, allowing them to progress more rapidly 

than AI-powered protective tools. Neural fuzzing employs neural networks to detect 

vulnerabilities in AI-powered systems, which enables threat actors to acquire 

knowledge from existing AI-powered systems (Jada, 2024).  

 

Explainable AI (XAI) - knowing why and how algorithms make their decisions, is critical 

in CS, and current AI algorithms lack that transparency, which erodes and questions 

their trustworthiness (Tiwari, 2023). The matter is sometimes exacerbated by the 

intricate neural networks and DL algorithms, as they function in ways that are not easily 

comprehensible to humans. XAI allows security analysts to comprehend the reasoning 

behind AI-generated alerts, suggestions, or choices by revealing the inner workings of 

AI algorithms (Sontan, 2024). Considerable progress has been achieved in the domain 



 45

of XAI, as researchers have devised strategies and procedures to augment the 

transparency of AI systems (Tiwari, 2023). 

 

The deployment of AI in cybersecurity is challenging. According to Aggarwal (2023), 

AI systems require extensive data to function effectively, and processing such volumes 

can be resource intensive. The risk of false alarms can undermine user trust in AI 

systems, and delayed responses to threats may compromise system effectiveness. AI 

models trained on biased or unrepresentative datasets have the potential to acquire 

and magnify existing biases, resulting in outcomes that are unjust or discriminatory. 

(Ferrara, 2023). Training data biases can arise from historical inequities, cultural 

prejudices, or sample biases, leading to biased predictions or choices by AI systems 

that disproportionately affect specific groups or individuals. It is crucial to prioritise 

establishing impartial data and model development to advance equitable and ethical 

AI. Methods such as dataset augmentation, bias-aware algorithms, and user feedback 

can be used to address the issue of bias in AI (Ferrara, 2023). 

 

There is no one-size-fits-all solution like traditional firewalls, anti-virus or anti-malware 

software (Cucu, 2019). Each AI solution must be tailored to a specific organisation 

using organisational-specific internal and external data (Raimundo, 2021). This 

requires greater financial and manpower commitment, and greater hardware and 

infrastructure commitment to implement. Whilst traditional network security tools such 

as firewalls and anti-virus software are universal, they are seen today as border-based 

protection (Xu, 2021). AI-driven NASs have substituted these traditional network 

protection mechanisms as they are faster and more efficient (Aliyari, 2021). 

 



 46

2.6.2 Organisational Challenges  

The future of AI evolution involves tackling emerging threats, scalability, deployment 

challenges, and organisational readiness, including necessary training for 

cybersecurity professionals.  

 

One of the main obstacles to multi-lateral and multi-vertical adoption of AI is 

infrastructure and hardware requirements, as significant CPU and manpower are 

needed, skills which are currently not present in the industry (Dawson, 2021). 

Implementing AI requires a range of specialised professionals, such as data scientists, 

data analysts, AI experts, machine learning specialists, developers, cybersecurity 

specialists, and project managers, each with different levels of technical expertise. 

(Attaran, 2018). This bars the high costs required for once-off implementation and 

maintenance (Wilkins, 2018). Without the manpower and budget, pilot testing and 

implementation could take years across an organisation. This excludes the starting 

point of systems, some of which are legacy and require a complete technological 

overhaul.  

 

Another challenge lies in the compatibility issues caused by the continued use of 

outdated systems, programming languages, and overall technological infrastructure in 

many organisations. Legacy systems fails to adequately support the requirements of 

AI and ML technologies. For instance, the analysis of vast amounts of complex data, 

a critical step in successful AI and ML deployment, is hindered by the lack of scalability 

offered by legacy databases and obsolete systems. Implementing AI solutions in 

organisations is not a simple task, as it often necessitates a complete overhaul of the 

technological infrastructure (Arasada, 2021). 



 47

The literature consistently highlights a recurring theme of insufficient high-quality, 

error-free, clean data availability. AI solutions rely on extensive datasets for training 

models and achieving accurate results. As a result, obtaining a large quantity of data 

is essential for training AI models effectively (Sun, 2022). Implementing a cyber-AI 

solution necessitates a more complex organisational data management process due 

to the diverse volumes and types of data stored, the speed at which data is 

accumulated, the need to maintain data confidentiality, and the constant requirement 

for additional data (Raimundo, 2021). This aspect is particularly crucial because the 

intelligence of AI solutions relies solely on the quality of the datasets used to train the 

models. 

 

2.6.3 Environmental Challenges  

There are no global, national, or industry-specific regulations to manage the 

enforcement of AI, let alone CS operations. The ethical and moral challenges lie with 

individuals and banks themselves in controlling data. The lack of regulation enables 

insider breaches and attacks for financial gain.  

 

2.7 Defining Perception Theory  

Andrew Demuth in Perception Theories (2013) Analyses the consequences of higher 

cognitive processes and its link to perception. What is perception? Where is the origin 

point of perception? What informs what we perceive and how we process the 

information we perceive? Perception Theory walks the bridge between general and 

cognitive psychology and philosophical epistemology and prompts one to explore the 

episteme and ontology of perception. The notion of perception is important to this 

paper because the core question is to understand the perception of AI in CS 

operations. As witnessed in the literature above, AI is anthropomorphic, it has a 



 48

scalable feature and operates through a centralised system that can easily proliferate 

the essence, DNA and fabric of an organisation. It also has the propensity to alter the 

ethos of a department and organisation. The perception of the effect AI has on the 

technological resources, on the intra and inter-organisational dynamics for the 

promotion of growth and well-being, as well as paying heed to regulations to protect 

the integrity of data conferred by customers, is crucial. While AI is in vogue and at the 

forefront of the digital revolution, the underside of the belly is dark and dangerous. 

Thus, the perception of AI needs to be an ongoing question to purview the landscape 

and determine its success or lack thereof. Are the challenges of AI to CS outweighing 

the benefits, and is there a non-AI solution? Are banks getting carried away by the 

current of AI? What is the social impact due to a mammoth disparity between AI-user 

organisations and non-AI user organisations? Is the continuous implementation of new 

AI features necessary? Does AI truly support CS operations or make operations more 

challenging to manage? Perception Theory will provide a frame of reference for 

exploring the mapping of how to answer these cutting-edge questions.  

 

The weightiest question in Perception studies is ‘What is the source of our cognition?’ 

The perspective is bifurcated into internalists, who propound a top-down approach and 

externalists, who espouse a bottom-up approach (Demuth, 2013). The former school 

and body of thought proclaim that knowledge, its sources and principles can be found 

within the subject itself, and cognition is but a mere discovery of a priori pieces of 

knowledge. The internalists are championed by Descartes and Plato, who believe that 

knowledge is about the collection of already acquired contents. All knowledge is pre-

existing and present; our perception and cognition bring realisation to what is (for 

example, the source of MA is nature). The second school looks at cognition through 



 49

an external sense, that being – experience. The externalists asserts that the mind is a 

blank sheet of paper (tabula rasa), and an external reality prints all knowledge. The 

externalists adopt a phenomenological stance in that they view cognition as a 

composition of social constructs created by personal, familial, cultural and socio-

economic conditions, and the on-going overlaying of experiences constantly alters 

one’s perception of their environment and world (Demuth, 2013, p. 15). Perception is 

the rationalisation of one’s personal experiences.  

 

 
Figure 2.3: Difference between Internalists and Externalists 

 
A question confronting authors and supporters of both schools asks the next question 

of how one can explain the awareness of a new reality in our consciousness. How 

does consciousness/the mind meet the world? The point that both schools agree on 

is the trustworthiness. The trustworthiness of a statement is maintained by the 

trustworthiness of the source/authority that postulates it (akin to a non-manipulative 

subject-matter expert). As Descartes articulated, ‘we must verify the knowledge 

principles themselves.’  The verification of testimonies takes three forms (Demuth, 

2013): 



 50

1. I verify data myself, but it may not be possible when one needs to test the data 

using an instrument. For example, using a microscope as opposed to optical 

vision can alter or skew information 

2. Testing the testimonies by mutual confrontation – Populus data testing on a 

large target sample. If most people share the same sentiment, it must be 

correct. However, the majority is not always correct 

3. Critically reassess the reliability and competence of each witness and their 

testimonies—In the judicial system, the verdict of the witnesses is important 

based on the background and sanctity of their truthfulness. This is tested by 

assessing how a sample witness arrives at his own testimony by verifying what 

he has already seen (his perception). Point three satisfies the paradox of one 

and two.  

 

Thus the synthesis of the internalist view (all knowledge is already pre-existing it is 

only rationalisation and awareness to it that is required) and the externalists view 

(one’s perception is based on personal experience and to each individual that will differ 

based on the matrix of social constructs he operates from) is a collective 

acknowledgement from a large body of people who all have experience on a subject 

matter and can verify data based on their own internal and external sense of 

perception.  

 

Now that the horizontal paradigm has been established, the vertical paradigm will 

examine the top-down and bottom-up approaches to perception.  

 



 51

2.7.2 Cognition, Science and Perception Flow   

Philosophers were the first to explore perception. As centuries passed and the 

Scientific Revolution took hold, there was a need to methodologically test every idea 

and perception to assess if it fell within reality. Philosophy provided the fodder for the 

inquisitive nature of science. “Philosophy enriched science not only via concepts and 

ideas, but also through distinctive methodology and ways of thinking. As good 

examples, we can mention the introspective method, the phenomenological reduction, 

and the description of phenomenological experience.” (Demuth, 2013, p. 20).  

 

According to the direction of data acquisition and processing, information can be 

positioned into a ‘top-up’ or ‘bottom-down’ approach. On the one hand, the bottom-up 

approach (herewith the neurological-cognitive stance will not be looked at) views data 

from a distance first and, upon closer inspection and engagement, can see and 

experience the complexity of processes. On the other hand, the vantage point for the 

top-down approach (again, the neurological-cognitive stance will not be looked at) is 

feeling. There is an inner sight and inner feeling about a subject matter that probes the 

exploration into determining action. It is top-down because the feeling appears as a 

thought in the mind and trickles into bodily action via the exploratory process. The 

essence of the top-down approach is tied to the view that one needs to have 

knowledge and experience in a field to help organise the cognitive contents (Demuth, 

2013, p. 23).  

 



 52

 
Figure 2.4: Difference between PT’s Top-down Bottom-up Approach 

The internalist-externalist view and top-down bottom-up perspectives are significantly 

richer and detailed. While this paper wishes to explore the exactaties of the ethos on 

a grander scale due to constraints, the door will be left open for future explorations 

and determinations. In what follows is the adaptation of Perception Studies to AI and 

CS.  

 

2.8 AI-CS and PT – A Synthesis 

The author's personal experience will act as the basis of the investigative eye in this 

synthesis/processing juncture, with its zenith being explored in semi-structured 

interviews and presented in the findings (Chapter 4).  

 

From the outset, there is an already systematic placement of the subject on both the 

internalist and externalist spectrum. Regarding the former, the 

author/researcher/subject has a priori experience in CS operations in the banking 



 53

sector and has first-hand experience in witnessing the evolution of AI tools and 

software from traditional firewall and malware tools to AAI and GenAI in the CS realm. 

The questioning of efficacy emerges from the ongoing exposure to the dynamics and 

effects of AI in CS. Positioned on the tabula rasa part of the externalists' view, the 

experience and almost AI-saturation in CS operations warranted the questioning of 

‘what else’ and ‘what next’. Apart from the anthropomorphic nature of AI, which is in 

the infant stage of its revolution, what other technological pursuits can be created, 

gauged and utilised to prevent an AI burn-out that has deep-seated ramifications for 

organisations and society as a whole given the lack of regulation in the industry? 

However, there is a leaning towards the externalists' position given the basis of 

perception questioning that emerged from tenure and experience in the industry.  

 

 
Figure 2.5: Locating the subject on Perception Theory's Quadrants 

 

Perception according to the horizontal paradigm is that experience within AI is a critical 

necessity when deciding what features to implement or not implement. The resources 

– personnel, infrastructure, and financial - are all needed to create a supporting 



 54

environment for AI implementation, without which the system has a way of failing the 

organisation. This is specific to CS operations within SA banks and broadly applicable.  

 

Across the vertical axis of bottom-up and top-down, the author/researcher/subject 

aligned predominantly towards the top-down approach as the genesis of the thought-

piece is the semi-mal-adaptation of AI in CS due to organisational pressures. 

Interestingly, the proximity to aligning with the top-down approach emerged from the 

bottom-up perspective. An analogy will be used: in the bottom-up perspective, a 

subject views an object from a distance, and with closer and closer engagement, the 

nuances and complexities can be seen and experienced (Demuth, 2013). The 

example used will be that of an apple. An apple, based on the outside covering, can 

be understood based on form, colour and size. Its texture can be gauged through 

touch. However, to the onlooker, the contents of the inside of the apple are unknown. 

The subject can bite into the apple and taste it, a step closer to understanding the 

apple – the taste, the texture, the juice, and so forth- but still, the contents are unknown 

to him. Upon the slicing the apple in half, the subject sees the core and the 

penultimation in the experience is the realisation that the apple's seed contains many 

apple trees. The researcher’s engagement with AI and CS affords him the 

trustworthiness, credibility and validity.  

 

Secondly, the vertical and horizontal axes are currently seen as static. To make them 

dynamic, they are viewed in 3D. This 3d form and credibility of perception of AI 

improving the efficacy of CS operations is tested by taking a subset of personnel, like 

the researcher, and testing their perception. As discussed in Chapter 3, 12 samples 

with varying levels of experience (2-28 years) were used in semi-structured interviews 



 55

to gauge their perception, sentiment and outlook of the efficacy of AI in CS operations 

in banks in SA.  

 

The articulation of the totality of the impetus and sentiment of the paper related to the 

ongoing effectiveness of AI in CS operations in SA banks can be experienced in the 

following quotation: “Where in the world is a metaphysical subject to be found? You 

say here it is just as with the eye and the field of vision. But you do not really see the 

eye. And nothing in the field of vision allows the conclusion that it is seen by an eye.” 

- Ludwig Witgenstein, Tractatus Logico–philosophicus, sentence 5.633 (2010). 

 

2.9 Literature Review Findings Viz Research Questions  

Research Question1: To what extent can artificial intelligence serve as an effective 

solution for eradicating and preventing cybercrime within the South African banking 

sector? From the literature, AI is not seen as a turn-key solution and is not viewed and 

used to eradicate cybercrime wholly. It is a mitigatory, reactive response that is 

anthropomorphic , and as new threats and attacks arise from complex, sophisticated 

attackers and software, the CS teams are in a position of response. The use of AI is 

to advance the banks position, and defence strategies are developed based on the 

views of potential threats and when threats and attacks occur in real-time.  

 

Research Question 2: What are the factors that contribute to the decrease of the CS 

staff complement in banks to improve ‘organisational effectiveness’? There is an 

overwhelming perception that the introduction of AI results in an almost immediate lay-

off of staff, be it within or outside of CS teams.  

 



 56

Research Question 3: What evaluation criteria or assessment frameworks are 

employed by banks to assess the effectiveness of AI-CS capabilities in reducing 

cybercrime through preventative measures? The body of literature is exceptionally thin 

and infantile in nature on CS operations and AI and CS. A lexicon has just been 

created, and most studies (academic and technical) do not cover internal assessment 

matrices, banks, and organisations are using vis-à-vis to defend software and tools to 

monitor output. Presently no template/dashboard exists that is publicly available. This 

is not to say internal CS operating teams do not have one, but based on the available 

literature, none exists as AI tools are custom and augmented to a team’s needs and 

scale requirements.  

 

 

 

2.10 Conclusion  

The TOE framework and Perception Theories created a scaffolding to identify and 

critically assess the body of literature of AI in CS operations in South African banks. 

Based on the analysis thus far, the approach and findings were both novel. Firstly, 

through the TOE framework, it was revealed that AI is generally viewed as a single 

solution that can either be integrated and adapted or not, and that the latest solution 

is the correct solution, precisely because of the perception that the latest update and 

upgrade is the best. However, the anthropomorphic nature of AI revealed that, 

according to an organisation’s needs, certain elements, features and infrastructures 

can be used and others not utilised at all. A legal firm may need simple protections 

and not the full GenAI feature. The Organisational modality revealed the lack of expert 

personnel resources needed to manage the system, and the scope for human error is 



 57

large; therefore, the quality of CS officers is of utmost importance. The environmental 

component revealed that the speed and velocity of the absorption of AI in SA banks 

far supersedes the time it takes to pass regulation. This places banks and CS officers 

at the helm of safeguarding and defending highly personal data of millions of 

individuals. The amount of power banks has accrued is becoming unprecedented and 

borderline dangerous. Via the literary discourse, it became clear that once an 

organisation starts on the path to advanced AI features, there is a point of no return. 

This, to a degree struck off and negated the notion that perception even counts, for 

once on the single road path, one must stay on it till the end.   

 

Secondly, through the analysis of Perception Theories and locating where and how 

perception forms, the process and outcome was one of empowerment. It located the 

author/researcher/subject as the focal point of the analysis, given his tenure and 

experience of working with AI in CS operations within South African banks, and 

positioned him on the quadrant of ‘externalist-top-down’. This view becomes a point 

of empowerment for the questioning of efficacy of AI in CS operations itself will prevent 

a herd-mentality of ongoing AI absorption and implementation, and place cognitive 

checks and balances in place before future AI uptake.  

  



 58

Chapter 3: Research Methodology  

3.1 Introduction 

The purpose of this research is to explore the impact AI has on improving the 

operational efficiency of cybersecurity in banks. While adopting and implementing AI 

across banks is rampant, does AI actually decrease the CS team’s workload and, 

based on machine-based learning and predictability, engage in smart learning to 

prevent attacks, or is it merely a tool/resource for responding to threats?  

 

Two things are occurring concurrently. Firstly, there is not a large body of literature 

focused on the impact of AI in CS, enabling this research to fill the gap and guide 

potential users in deciding whether AI benefits them. Secondly, given the devastating 

impact leaked information could have on institutions, no prevalent data is existing that 

banks and other financial institutions are willing to readily share severely limiting the 

knowledge base of the true impact of AI and AAI on CS operations. Knowledge of AI 

and CS, both in operating and academic circles, is both slim and infantile in nature. 

Given the highly specialized nature of roles of CS officers and any operating tools 

being public knowledge, it will give fodder to cyber criminals hence the data available 

is scarce. To mitigate this challenge a qualitative approach was adopted. 

 

A qualitative approach guided by interpretivism as the research philosophy and the 

adoption of the Phenomenological Approach, combined with the researchers’ personal 

experience, allowed for highly applicable questions and ease of questioning to security 

officers in banks in SA. A sample of 12 participants were interviewed sequentially, 

allowing for healthy engagement and dialogue and creating space for nuanced 



 59

information to flow. Given the very thin body of literature on the subject matter, the 

findings of the analysis will inform the creation of a new model/theory that CS officers 

across sectors can use as a litmus test and an implementation framework to ascertain 

if AI is the most viable solution for their platforms and goals. The process is mapped 

below: 

 

 
Figure 3.1: Map of Research Methodology 

 

3.2 Research Philosophy  

In a study by Kingsley Ofosu-Ampong (2024), he scoped the exact amount of literature 

available in academic and financial services domains concerning data on AI. Of the 

content published and produced over three years (2020-2023), only 14 per cent was 

qualitative. By adopting a qualitative approach, the paper will contribute significantly 

to the infantile nature of the topic itself and drastically aid in the body of literature and 

nature of findings in both academic and digital circles. Apart from the novel nature of 

the topic and given the key limitation of privacy of data to conduct a quantitative study, 

the author has strategically adopted a qualitative approach.  



 60

A qualitative approach aims to facilitate the acquisition of understanding through the 

experiences of others. Qualitative research designs possess several important 

qualities. These include a focus on comprehending meaning, utilising non-numerical 

data, employing inductive analysis, including subjectivity, utilising small sample sizes, 

maintaining flexibility, gathering rich and thorough data, and placing emphasis on 

context (Sharma, 2023). This method was selected as the research investigates the 

perceived impact of AI on cybersecurity operations in the South African banking sector.  

 

According to Ghafar (2024), qualitative research data offers several advantages: It is 

a detailed and comprehensive description of the participants ideas, beliefs, and 

experiences; It most accurately represents the human experience within specific 

circumstances; It provides various viewpoints, research methodologies, and tools to 

understand an individual’s experiences and; It gives an accurate and thorough 

assessment of a subject because they enable participants to ascertain which study 

aspects hold the highest significance for them. Interpretivism is the research 

philosophy adopted to answer the qualitative paradigm's core and subsidiary research 

questions.  

 

Four main research philosophies exist: positivism, interpretivism, critical realism and 

pragmatism. Interpretivism was the most appropriate given that the research 

methodology is qualitative, with the main approach being semi-structured interviews 

for a particular target and sample group (Jain, 2023). The results are naturally 

subjective in nature but given the infantile schematic of qualitative data on AI in general 

and AAI-come-CS, a qualitative view will offer an ontological view that a purely 

positivistic approach would be unable to generate. Interpretivism recognises that 



 61

“humans are different from physical phenomena because they create meaning” 

(Saunders, 2023, p. 150) and that reality is a navigatory matrix composed of the 

intersections of social constructs, the outcomes of which shape experiences, 

perceptions and interactions.  

 

By adopting interpretivism, which is inductive, the study explored the true lived 

experience of people with knowledge of specific phenomena. It uncovered the 

dynamic layers that not only constitute perceptions of AI and CS but also promote 

healthy cognition towards the approaches taken to the betterment of roles and 

resources. The aim was not to blanket participants' responses but to understand the 

rationale and perception behind the participant's interactions and engagements with 

AI, given their roles in CS.   

 

3.3 Research Assumptions  

The ontological nature underlying interpretivism navigates the intersectionality of 

constructs and the outcomes produced because of. The nature of semi-structured 

interviews combined with the authors personal experience with AAI in CS enabled a 

healthy dialogue that unpacked the nuances of how the theological nature of AI is 

embedded into the ethos of banks, as well as AI being multifaceted in nature and a 

blanket answer to its perception and efficacy will not do justice to its impact in 

protection and digital development.  

 

As will be seen, depending on one's tenure within AI and one’s roles within CS, the 

outlooks of the participants have a natural similarity and natural variance. Interestingly, 

the outcomes tallied with the AI themes of Ampong (2024); that is, AI and AAI fall into 



 62

the typologies of technological issues, contextual knowledge co-creation issues, 

conceptualisation, and domains and applications. The typologies correspond with the 

interpretative, exploratory nature of the study, which recognises that AI in CS will have 

multiple constructs. However, the constructs exist within a system, and the nature of 

all ecosystems is the natural development of an order and hierarchy, which also has 

an unpredictive angle. The simultaneous interplay of agency within structure guided 

the qualitative approach and design.  

 

3.4 Research Design  

There is a plethora of qualitative research designs, each with its unique characteristics 

and purposes. According to Among (2024), there are presently no frameworks on CS 

and AI and AAI given the generic nature of AI as well as the scant data produced on 

AAI and CS (There are general qualitative and quantitative AI frameworks such as 

TAM; Fuzzy logic; Self-determination Theory; Critical Theory and Antromorphism 

Theory, but they are not fully applicable to CS). As such, a phenomenological research 

design will be utilised.  

 

According to Jain (2023, p. 3) phenomenological research “Aims to understand the 

essence and meaning of human experiences related to a particular phenomenon. 

Researchers explore participants’ subjective experiences through in-depth interviews 

or observations to uncover the underlying structures and patterns of their lived 

experiences.” This approach successfully marries with the exploratory interpretive 

guiding philosophy. Data was gathered via comprehensive semi-structured interviews 

with individuals who are experiencing the phen