INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025  
GenerativeAI and Privacy-Preserving Big DataAnalytic in Cloud  
Environments withAI Agents  
Akanksha Shukla, Dr. Rohit Kumar  
Haridwar University, Roorkee  
DOI: https://doi.org/10.51583/IJLTEMAS.2025.1411000124  
Received: 08 December 2025; Accepted: 15 December 2025; Published: 25 December 2025  
ABSTRACT  
While generative artificial intelligence (GenAI) technologies are revolutionising content production, they also  
pose serious privacy and data security issues. The potential of privacy violations, biases, and cyberattacks rises  
as these models process large datasets, many of which contain sensitive or private data. These issues are  
examined in this book, especially in important fields like cybersecurity, healthcare, and finance. The potential  
for GenAI models to reproduce or infer sensitive data from training datasets is a major problem that raises ethical  
and intellectual property issues. Data protection techniques like encryption, tokenisation, and anonymisation are  
crucial to reducing these dangers. This study assesses the efficacy of these techniques by looking at how they  
affect the functional performance and privacy risk reduction of GenAI systems. It evaluates the impact of  
tokenisation and anonymisation on a state-of-the-art large language model (LLM) through experimental analysis.  
Empirical results offer insights into the trade-offs between protecting model performance and data privacy using  
open-source tools such as Microsoft Presidio. The goal of the research is to help create safe and morally sound  
GenAI applications, making sure that advancements in AI are in line with data security guidelines while  
preserving accuracy and efficiency in practical applications.  
Keywords: Privacy-Preserving, Big Data,AI-Driven,Cloud and Techniques.  
INTRODUCTION  
GenAI has exploded in the previous two years, with exponential growth. It can be used to create realistic,  
imaginative text, graphics, and other types of data, including music. This suggests that there will be substantial  
uses for this advancement in GenAI across a range of sectors, such as marketing, healthcare, finance, and  
entertainment. But such quick expansion brings up important issues about information security and data privacy.  
First, in order to train effectively, generative AI requires a vast amount of data [1]. Additionally, when sensitive  
personal data is included in the data, this naturally raises the danger of leakage. Any contact with the GenAI  
system could add to a dataset that contains personally identifiable information; therefore, if appropriate  
anonymisation or data protection is not in place, that dataset would become vulnerable. The lack of openness in  
data collecting, storage, and utilisation continues to be one of the primary issues. Most of the time, end users are  
unaware of the maximum amount of data that can be used, particularly when that data is shared or processed by  
an outside service provider. Because these outside contractors could not adhere to the same stringent privacy  
regulations as internal services and might utilise the data more frequently, outsourcing can raise the security risk.  
User data, for instance, could be utilised for reasons other than data. Apart from raising concerns about who  
owns and controls personal data when it is sent to these platforms, this seriously infringes on a person's right to  
privacy. Unintentionally disclosing intellectual property is the other significant risk [2].  
Sensitive corporate data may be accidentally accessed or shared as a result of the model's training process  
absorbing proprietary or secret information that is supplied to it by individuals and businesses. Because so many  
GenAI platforms store data in cloud environments, there is an increased chance that private data could be stolen,  
intercepted, or used in other ways by cybercriminals. The fact that AI models are black-boxes makes it difficult  
to understand or even track decisions or internal data processing, which increases these dangers [4].  
www.ijltemas.in  
Page 1334  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025  
Figure 1. Proposed Privacy-Privacy workflow in AI [3]  
In addition to creating accountability issues, this opaqueness makes it extremely difficult to ensure adherence to  
privacy rules and regulations, such as the GDPR. Given these grave worries, it goes without saying that GenAI  
platforms will seriously jeopardise user privacy, data security, and intellectual property issues if they are not  
protected, with far-reaching consequences for both people and organisations.The following are some of the  
report's key goals: Examining the threat landscape related to GenAI with an emphasis on the dangers to  
intellectual property, privacy, and security.  
1. To investigate several approaches to risk reduction and data security that can be used for the responsible  
development and application of GenAI.  
2. It explains a particular data tokenisation project, including its implementation, outcomes, and constraints, in  
order to thoroughly examine how data tokenisation is one potential way to improve data privacy in the  
context of GenAI.  
The rest of the following section are, in Section II literature survey related this research has been explained. In  
Section III, proposed methodology has been elaborated. In Section IV, results has been showed and discussed  
with conventional work. Finally, in Section V, conclude the proposed work.  
LITERATURE SURVEY  
In order to improve Cyber-Physical Systems (CPS) in the medical domain, [5] provides a comprehensive analysis  
of deep learning in conjunction with image categorisation. The authors highlight the vital role that deep learning  
plays in picture classification while deftly navigating the complexities of secure medical environments. By  
addressing the intersection of technology and healthcare, the research contributes to the evolving field of safe  
systems in an important area.  
In [6], examine the potential for future wireless communication with a focus on 6G. Their study demonstrates  
the advantages of integrating blockchain and artificial intelligence, offering insights into how these two  
technologies could cooperate to enhance security and privacy in the emerging 6G landscape. By providing both  
a theoretical foundation and real-world applications, this study significantly adds to the conversation about the  
security issues with growing wireless networks.  
In [7] offers a comprehensive examination of critical security for Internet of Things (IoT) networks,  
encompassing blockchain, AI, and conventional methods. The paper is a priceless resource for academics and  
business people alike by analysing the numerous security concerns associated with IoT. To broaden the  
discussion and offer a thorough road map for understanding and addressing security concerns in the quickly  
evolving IoT network environment, a number of security paradigms are included.  
www.ijltemas.in  
Page 1335  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025  
An in-depth analysis of the convergence of distributed ledger technology (DLT) with artificial intelligence (AI)  
can be found in [8]. Their paper provides a current evaluation of the state of this convergence, highlighting its  
primary challenges and outlining potential directions. This survey is a helpful resource for understanding the  
evolving landscape at the nexus of DLT and AI. It was published in IEEE Access. Both academics and business  
professionals can benefit from its insights.  
Reliable and privacy-preserving federated deep learning is emphasised in [9], which contributes to the corpus of  
work in the context of the Industrial Internet of Things (IIoT). Their work addresses the critical need for robust  
security and privacy protections in IIoT designs. In addition to including federated deep learning, the proposed  
method prioritises privacy and trust preservation, taking into account the particular requirements of the industrial  
context.  
Figure 2. Security Risk Distribution on Generative AI Platforms [10]  
Table 1. Comparative Analysis related to Cloud based AI gent techniques for privacy and Security [11]  
References Year  
Title  
Focus  
Contribution  
Provides a comprehensive survey  
of privacy and security risks in  
Privacy  
Implications of Cloud-Based  
AI Services: A Survey  
and  
Security  
cloud-based  
introducing  
categorize  
AI  
taxonomy  
risks  
services,  
Cloud-Based  
AI Services  
[12]  
[13]  
2024  
a
to  
and  
these  
discussing defenses for both model  
providers and consumers.  
Proposes  
a
secure  
cloud  
architecture that ensures data,  
logic, and computation remain  
secure during transit, use, and at  
Towards  
Confidential  
Computing: A Secure Cloud Secure Cloud  
Architecture for Big Data Architecture  
Analytics and AI  
2023  
2023  
rest,  
addressing  
concerns  
in  
biomedical research and other  
sensitive fields.  
Explores the integration of AI and  
blockchain  
enhance  
technologies  
privacy, discussing  
to  
An Overview of AI and AI  
Blockchain Integration  
Privacy-Preserving  
and  
for Blockchain  
Integration  
[14]  
applications in data encryption, de-  
identification,  
and  
multi-tier  
distributed ledgers.  
www.ijltemas.in  
Page 1336  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025  
Provides a comprehensive survey  
Privacy-Preserving Data in  
IoT-based  
Cloud  
Systems  
of privacy issues in IoT and cloud  
systems, highlighting the role of AI  
in dynamic anonymization and  
secure data sharing.  
IoT-based Cloud Systems: A  
Comprehensive Survey with  
AI Integration  
[15]  
[16]  
[17]  
2024  
2024  
2023  
Surveys privacy concerns specific  
to  
generative  
AI  
models,  
Generative AI Model Privacy: Generative  
discussing potential risks and  
mitigation strategies to protect  
sensitive information.  
A Survey  
AI Privacy  
Analyzes various security and  
Data privacy solutions for big data  
and Privacy-Preserving Big Analytics in analytics in cloud environments,  
Systematic Survey: Secure Big  
Data Analytics in Cloud  
Cloud  
focusing on secure access control,  
data storage, and private learning.  
Discusses security and privacy  
challenges  
generative data in AI-generated  
Data Security content (AIGC), providing insights  
into current solutions and future  
directions.  
associated  
with  
Security  
Generative Data in AIGC: A  
Survey  
and  
Privacy  
on  
Generative  
[18]  
2024  
These models serve as the foundation for training machine learning models, enhancing datasets, and protecting  
data privacy since they tackle the issues of data imbalance, privacy, and scarcity.  
PROPOSED METHODOLOGY  
This suggested approach evaluates and selects a single anonymisation technique to be used for extensive LLMs  
and assesses how effectively it performs in a variety of scenarios involving different input types and personally  
identifiable information (PIIs). An detailed literature review and exploratory research that concentrate on the  
current state of anonymisation techniques and tools are the first steps in the process. The top-ranked open-source  
anonymizers are determined by consulting a variety of sources, including technical papers, industry publications,  
and scholarly studies.  
The changes of community's recognition, integration potential, generative AI support, and tool creators'  
trustworthiness. The strengths and limitations of each of these tools will be compared with this selection criterion.  
The instrument with the best ratio of strengths to shortcomings will be chosen for additional examination.  
Figure 3. Proposed GAN workflow  
www.ijltemas.in  
Page 1337  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025  
After selecting an anonymisation tool, the next step is to establish an experimental scenario that will be used to  
evaluate the tool's functionality. This would entail defining the kinds of data that need to be anonymised and  
creating the experiment's architecture using a range of PII and additional input formats. The LangChain  
framework or a comparable tool will be used to achieve anonymisation, and performance measurements will  
include processing speed, anonymisation correctness, and effect on LLM comprehension.  
The ROUGE/BLEU [19] anonymisation quality ratings, LLM-based evaluations of the model's comprehension  
for anonymised versus non-anonymized input, and human evaluations to offer qualitative insights into the tool's  
efficacy will be the evaluation metrics. These findings will be examined in a comparative analysis of the  
anonymisation tool's diverse performances across multiple scenarios, highlighting its versatility in handling  
different input types and PII types.  
The last step will include an evaluation of the effectiveness of the selected anonymisation tool as well as  
recommendations for how to make it better or different. Additionally, it will discuss potential directions for  
future study as well as practical application based on findings. A more structured method of choosing a system  
for anonymisation ensures that enough data on how it operates in specific contexts is obtained.  
Transparency and Data Minimisation  
By collecting and processing just the necessary data, data minimisation plays a crucial role in lowering the  
likelihood of a breach. At the very least, handling less amounts of data implies a lower chance of a massive data  
breach, which is very troublesome when it comes to sensitive or personally identifiable information. Conversely,  
transparency refers to informing consumers about the potential benefits of the information gathered about them,  
so they are fully aware of how AI processes data. The degree of openness will foster trust, which will motivate  
users to provide informed consent in order to fulfil their ethical duties [20].  
Protection of Data  
The foundation of data protection will be adaptive character AI security solutions, which will evolve over time  
in tandem with new threats and technical advancements. As a result, the process of identifying and screening  
enabling technologies must be especially cautious when it comes to the tools, libraries, and frameworks that are  
crucial to the development and use of AI. With a few notable exceptions, open-source technologies have become  
more prevalent in the development of AI systems [21].  
Verification  
Following the screening of the enabling technologies, application and infrastructure security will be the main  
focus. The majority of AI systems function in extremely intricate ecological settings, where infrastructure flaws  
potentially jeopardise their reliability. Effective security methods, such as MFA, data encryption, and RBAC,  
will be necessary to safeguard the AI systems themselves [22].  
Ongoing Observation  
Organisations must also keep an eye on AI-specific risks, which are distinct from conventional cybersecurity  
issues, in addition to safeguarding their infrastructure. Adversarial input, for example, is a sort of attack that  
targets AI systems exclusively. By making small changes to the input data, malicious actors might alter the  
appearance of the AI output. Data poisoning, in which training datasets are tainted to produce inaccurate AI  
predictions and behaviours, is the other major hazard [23].  
Handling Vulnerabilities  
The institutionalisation of policies pertaining to vulnerability management will be the other key focus of AI  
security. This relates to routine risk assessments and vulnerability scanning for potential attempts to take  
advantage of system flaws. Here is where a company may guarantee weaknesses by continuing to take a  
proactive stance in identifying threats. These are vulnerabilities that can be exploited before they are exploited  
www.ijltemas.in  
Page 1338  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025  
if they are found and repaired. In addition, prompt incident response may be crucial to minimising security  
vulnerabilities as soon as feasible [25].  
As a result, the following strategies are suggested for the fundamentals of GenAI: risk management, transparency,  
security, and data minimisation. Every company should have a comprehensive AI security plan that anticipates  
how technologies will always make it stronger and more secure, both at the data and infrastructure levels, which  
are the foundation of AI systems. In this regard, ongoing vulnerability monitoring and management, along with  
ethical standards, will help the company significantly lower the risks connected with AI while creating more  
safe, trustworthy, and socially responsible systems.  
RESULTS AND DISCUSSION  
Randomisation, the process of adding noise to data, is frequently done via a probability distribution.  
Randomisation is applied in sentiment analysis and surveys. Randomisation does not require knowledge of other  
entries in the data. It can be applied to the stages of data collecting and preparation. There is no anonymization-  
related overhead with randomisation. However, because of time complexity and data utility, randomisation is  
not practical for large datasets, as demonstrated by our experiment, which is described below [26].  
More Mappers and Reducers were used as the amount of data increased. There was a considerable difference  
between the results before and after randomisation. Randomisation has little effect on a small number of outlier  
records, which are vulnerable to adversarial assault. When it comes to attribute sharing, randomisation might not  
be the best way to protect privacy because data utility is not valued when privacy is sacrificed.  
Table 3: After utilising age and zip code anonymization [27]  
Sr.no.  
Zip  
262  
362  
414  
536  
458  
Disease  
Age  
1
1
2
3
4
5
Cardiac problem  
Cardiac problem  
Cardiac problem  
Skin allergy  
3
2
>50  
>50  
Cardiac problem  
Table 3: T closeness privacy protection method [28]  
Sr.no.  
Age Record  
Zip Record Medical Record  
Salary Record  
5463  
1
2
3
4
5
6
5
263  
363  
424  
537  
459  
378  
Flu  
10  
Cardiac problem  
Skin allergy  
Cancer  
6352  
15  
7246  
>50  
>60  
>70  
8157  
Cardiac problem  
Skin allergy  
9463  
4681  
www.ijltemas.in  
Page 1339  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025  
Figure 4. T closeness privacy protection method  
The dataset provided contains information on a wide range of individuals who are identified by their serial  
numbers (Sno) and reside in different zip codes (Zip). Age, declared income, and any related medical conditions  
are used to identify each individual. The wage data is one significant feature of this dataset that adds a fresh  
viewpoint to the investigation. A range of ages are represented by patients with serial numbers 1 through 6, with  
an emphasis on those who are over 50 (referred to as ">50"). Interestingly, individuals in this age group have  
been diagnosed with a variety of diseases, including cancer, heart problems, and the flu.  
The dataset illustrates the potential relationship between the prevalence of several illnesses, age, and income.  
Patients with heart problems report salaries ranging from 5463 to 9463, indicating a range of income levels  
within this health category. Similarly, there is a variety in the reported salaries of those with skin allergies or  
cancer diagnoses. This dataset offers an opportunity to investigate the relationships among age, socioeconomic  
position, and the likelihood of specific health issues. Healthcare professionals and policymakers may find it  
crucial to understand these links in order to develop targeted interventions and healthcare policies that take into  
account the complex nature of health disparities within this group.  
CONCLUSION  
Conclusively, this proposed effort highlights the complex interplay between the urgent need for strong data  
protection measures and the transformational promise of generative AI. The study illustrates the potential and  
difficulties presented by this quickly developing technology by following the development of GenAI throughout  
time and examining its present capabilities, constraints, and security threats. Although GenAI has enormous  
advantages in terms of automation and content creation, it also comes with serious concerns, such as false  
information, privacy violations, and cyberthreats. Therefore, in order to guarantee ethical use and minimise  
potential harm, GenAI must be developed and regulated responsibly. In order to create a future where GenAI  
can flourish while maintaining data security and privacy, more research and proactive governance will be  
necessary.  
REFERENCES  
1. Chen Y, et al. Blockchain-based medical records secure storage and medical service framework. J Med  
Syst. 2019;43:1–9.  
2. Mayer AH, da Costa CA, Righi RDR. Electronic health records in a blockchain: a systematic review.  
Health Inf J. 2020;26(2):1273–88.  
3. Ghadi YY, et al. The role of blockchain to secure internet of medical things. Sci Rep. 2024;14(1):18422.  
4. Ghadi YY, Shah SFA, Mazhar T, Shahzad T, Ouahada K, Hamam H. Enhancing patient healthcare  
with mobile edge computing and 5G: challenges and solutions for secure online health tools. J Cloud  
Comput. 2024;13(1):93.  
5. Saranya R, Murugan A. A systematic review of enabling blockchain in healthcare system: Analysis,  
current status, challenges and future direction. Mater Today Proc. 2023;80:3010–5.  
www.ijltemas.in  
Page 1340  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025  
6. Andrew J, et al. Blockchain for healthcare systems: architecture, security challenges, trends and future  
directions. J Netw Comput Appl. 2023;215: 103633.  
7. Sujan MA, Looking at the safety of ai from a systems perspective: two healthcare examples, in safety  
in the digital age: sociotechnical perspectives on algorithms and machine learning. 2023, Springer  
Nature Switzerland Cham. p. 79–90. .  
8. Wehkamp K, Krawczak M, Schreiber S. The quality and utility of artifcial intelligence in patient care.  
Dtsch Arztebl Int. 2023;120(27–28):463.  
9. Mondal H, Mondal S, Singla RK, Artifcial Intelligence in Rural Health in Developing Countries, in  
Artifcial Intelligence in Medical Virology. 2023, Springer. p. 37-48  
10. Zuhair V, et al. Exploring the impact of artifcial intelligence on global health and enhancing healthcare  
in developing nations. J Primary Care Commun Health. 2024;15:21501319241245850.  
11. Poalelungi DG, et al. Advancing patient care: how artifcial intelligence is transforming healthcare. J  
Personal Med. 2023;13(8):1214.  
12. Taherdoost H, Machine learning algorithms: features and applications, in Encyclopedia of Data  
Science and Machine Learning. 2023, IGI Global. 938–960.  
13. Mirjalili S, Gandomi AH. Comprehensive metaheuristics: algorithms and applications. Amsterdam:  
Elsevier; 2023.  
14. Worden K, et al. Artifcial neural networks, in machine learning in modeling and simulation: methods  
and applications. Berlin: Springer; 2023. p. 85–119.  
15. Kasneci E, et al. ChatGPT for good? On opportunities and challenges of large language models for  
education. Learning Individual Dif. 2023;103: 102274.  
16. Chang Y, et al. A survey on evaluation of large language models. ACM Trans Intell Syst Technol.  
2024;15(3):1–45.  
17. Chaka C. Detecting AI content in responses generated by ChatGPT, YouChat, and Chatsonic: The case  
of fve AI content detection tools. J Appl Learning Teaching. 2023;6(2):12.  
18. Wu X, Duan R, Ni J. Unveiling security, privacy, and ethical concerns of ChatGPT. J Inf Intell.  
2024;2(2):102–15.  
19. Ma S, et al. “Are you sure?” Understanding the efects of human self-confdence calibration in ai-  
assisted decision making. in proceedings of the CHI conference on human factors in computing  
systems. 2024.  
20. Masood I, et al. A blockchain-based system for patient data privacy and security. Multimedia Tools  
Applications. 2024;83(21):60443–67.  
21. Salah M, Al Halbusi H, Abdelfattah F, May the force of text data analysis be with you: Unleashing the  
power of generative AI for social psychology research. Comput Hum Behav Artif Hum, 2023: 100006.  
22. Al-Hawawreh M, Aljuhani A, Jararweh Y. Chatgpt for cybersecurity: practical applications, challenges,  
and future directions. Clust Comput. 2023;26(6):3421–36.  
23. Yao Y, et al., A survey on large language model (llm) security and privacy: The good, the bad, and the  
ugly. High-Confdence Computing, 2024: p. 100211.  
24. Singh K, Chatterjee S, Mariani M, Applications of generative AI and future organizational  
performance: The mediating role of explorative and exploitative innovation and the moderating role  
of ethical dilemmas and environmental dynamism. 2024. 133: 103021.  
25. Tokayev K-J. Ethical implications of large language models a multidimensional exploration of societal,  
economic, and technical concerns. Int J Soc Anal. 2023;8(9):17–33.  
26. Usman M, Qamar U. Secure electronic medical records storage and sharing using blockchain  
technology. Proc Comput Sci. 2020;174:321–7.  
27. Vanathi J, G. SriPradha. BreakTheChain: A Proposed AI powered Mobile Application Framework to  
handle COVID-19 Pandemic. Alochana Chakra Journal. 9: 108–114.  
28. Yan L, et al. Practical and ethical challenges of large language models in education: A systematic  
scoping review. Br J Edu Technol. 2024;55(1):90–112.  
www.ijltemas.in  
Page 1341