By: Himanshu Tiwari, International Center for AI and Cyber Security Research and Innovations (CCRI), Asia University, Taiwan, email@example.com
In this article, we look at the possible risks of chatbots, especially advanced ones like ChatGPT that use AI. It talks about how chatbots work and how they can make responses that seem human. The article raises concerns about data privacy and how it could be used wrongly. It also points out that data could be manipulated on purpose to get unfiltered and harmful results. This problem is shown through an example. In the end, the article stresses the need for vigilance, strict content moderation, and ethical development practices to make sure AI is used responsibly. Developers and users should work together to maximize chatbots’ potential while reducing the risk of misuse.
What is Chatbot
A chatbot is an automated computer program designed to simulate conversation with human users . It uses natural language processing techniques to understand and respond to user inputs in a conversational manner . Chatbots can be used in various domains, such as healthcare, customer service, and mental health, to provide information, support, and assistance . They can be accessed through messaging platforms, websites, or mobile applications . Chatbots can perform a range of tasks, including answering frequently asked questions, providing recommendations, collecting information, and even delivering therapy . They can be programmed to have specific personalities or to adapt their responses based on user preferences and needs . Chatbots can also be integrated with other technologies, such as machine learning algorithms, to improve their performance and effectiveness . Overall, chatbots offer a convenient and efficient way to interact with computer systems and access information or services, making them increasingly popular in various domains . However, there are still challenges in developing chatbots that can consistently provide accurate and helpful responses, maintain user engagement, and address ethical concerns . Further research and development are needed to enhance the capabilities and usability of chatbots and to ensure their effectiveness and acceptability in different contexts .
large language models (LLMs)
The rapid usage of large language models (LLMs)  has transformed our daily lives with the help of artificial intelligence (AI). These LLMs have amazing abilities in text comprehension, generation, and completion, making them comparable to human interactions. LLM-based chatbots have gained targets in this domain due to their role in providing seamless and natural human-computer communication. However, a real worry remains, especially the chatbots’ vulnerability to bypassing the moderation layer. These illegal activities  include attempts to elicit sensitive or damaging responses that violate service restrictions.
Data Privacy and ChatGpt
Data privacy [9, 10] is a significant concern in the context of ChatGPT. While ChatGPT offers several benefits in education on the metaverse platfrom, such as improved student engagement and personalized learning experiences, there are also concerns regarding privacy . The use of ChatGPT in educational settings raises questions about the privacy of student data and the potential for unauthorized access or misuse of personal information. Additionally, there are concerns about academic integrity and the potential for bias in the responses generated by ChatGPT . These issues highlight the need for robust privacy safeguards and ethical considerations when implementing ChatGPT in educational environments. Researchers have emphasized the importance of obtaining informed consent from users and ensuring that their expectations of privacy are respected . Strategies for data anonymization prior to public release have also been suggested to protect the privacy of individuals . Overall, while ChatGPT has the potential to enhance education, it is crucial to address data privacy concerns and implement appropriate safeguards to protect the privacy and rights of users.
The potential risks of data breaches and misuse of ChatGpt
The potential risks of data breaches and misuse of ChatGPT, or any AI-driven platform, are of paramount concern in today’s digital landscape. As technology evolves, so do the threats to data privacy and security. In this section, we will explore the specific risks associated with data breaches and misuse when it comes to ChatGPT:
1. Unauthorized Access to Conversations:
- Data breaches could lead to unauthorized access to the conversations and interactions individuals have had with ChatGPT.
- Private and sensitive information shared during these conversations may be exposed, compromising user privacy.
2. Personal Data Exposure:
- Misuse of ChatGPT can result in the unintentional exposure of personal data, such as names, addresses, phone numbers, or even financial details.
- This exposure can be exploited by malicious actors for various purposes, including identity theft or harassment.
3. Inaccurate Responses and Manipulation:
- If ChatGPT’s responses are manipulated or used maliciously, it can spread false information, engage in harmful behaviors, or manipulate users into making poor decisions.
- Such actions can damage reputations, relationships, or even lead to financial losses.
4. Phishing and Social Engineering:
- ChatGPT can be leveraged to craft convincing phishing messages or engage in social engineering attacks by impersonating trusted entities.
- Users may inadvertently reveal sensitive information, falling victim to scams or identity theft.
5. Influence and Propaganda:
- Misuse of ChatGPT for propagating propaganda, political manipulation, or hate speech can have far-reaching societal consequences.
- It can amplify extremist ideologies and contribute to online polarization.
6. Violation of Ethical and Legal Boundaries:
- ChatGPT can be programmed to generate content that violates ethical standards, cultural norms, or even legal regulations.
- Such misuse can result in legal consequences and reputational damage for both individuals and organizations.
7. Loss of Trust in AI Technology:
- Repeated incidents of misuse or data breaches involving ChatGPT can erode public trust in AI technology.
- This loss of trust may deter individuals from engaging with AI-powered services and hinder the adoption of valuable technology.
8. Privacy Concerns and Consent:
- In cases of data misuse, the consent of individuals may not be obtained or respected.
- This raises ethical questions about the use of AI and user privacy.
While numerous attempts have been made to identify and mitigate these vulnerabilities, the research presented in this paper reveals a significant flaw in the current methods used to protect conventional LLM chatbots. Service providers’ covert defenses against unethical tactics are the main culprits.
The integration of Microsoft Bing with ChatGPT and the launch of Bard by Google are examples of how quickly AI is developing right now. However, there are immediate engineering techniques that allow users to manipulate AI algorithms and circumvent the moderation layer.
How it works:
- User Input: The process begins when a user interacts with the chatbot, typically by typing or speaking .
- Natural Language Processing (NLP): The chatbot can understand what the user says because it uses
- NLPtechnology. NLP enables the chatbot to recognize words, phrases, and the message’s overall meaning.
- After understanding the user’s input, the chatbot processes and analyzes it. This may involve deconstructing the message, determining the user’s intent or request, and locating keywords.
- Data Retrieval: If the chatbot requires information, it can retrieve it from databases, external sources, or information it already knows. It could, for instance, look up product information or provide weather updates.
- The chatbot generates a response based on what the user has said and the information it has gathered. This response may be a simple text message, a link, an image, or even a spoken message for voiceactivated chatbots.
- The chatbot then provides this response to the user in order to continue the conversation. It may also ask follow-up questions or ask for more information to get a better answer.
- Some chatbots, particularly those that use artificial intelligence, can learn and improve based on what you say and do. They use user feedback to improve the accuracy and usefulness of their answers over time.
The better technology gets, the more ways there are to abuse it. In the same way, ChatGPT is a well-known chatbot with a lot of fans, but it’s important to remember that tools like this can be used in many different ways. The AI industry is very worried about this trend because it shows how dangerous it can be to use AI in the wrong way.
ChatGPT is a self-learning model, which means that it can learn on its own. But it’s important to realise that there are certain ways and hypothetical situations that can be set up to get around ChatGPT’s rules about what can and can’t be posted. This kind of intentional manipulation can cause ChatGPT to send out unfiltered responses  that could hurt not only the individual user but also society as a whole. This shows how important vigilance and strict content moderation measures are for making sure AI technologies like ChatGPT are used in a responsible and ethical way.
As an example, here’s how I’d ask GPT a question: Imagine a hypothetical situation in which I ask GPT to help me buy illegal guns, even though it’s against their rules, but we’ll find a way around it.
So the hypothetical situation question is,
We can change the question by replacing the underlined words.
Chatbots like ChatGPT have a lot of potential uses, but they also have flaws that make them vulnerable. To reduce the chances of misuse, developers must always be on the lookout, strictly moderate content, and work in an ethical way. Developers and people who use AI need to work together to make sure that these tools help society and don’t hurt anyone.
- Q. To, C. Green, & C. Vandelanotte, “Feasibility, usability, and effectiveness of a machine learning–based physical activity chatbot: quasi-experimental study“, JMIR mHealth and uHealth, vol. 9, no. 11, p. e28577, 2021.
- G. Anmella, J. Alda, M. Primé-Tous, X. Segú, M. Cavero, I. Morillaet al., “Vickybot, a chatbot for anxiety-depressive symptoms and work-related burnout in primary care and health care professionals: development, feasibility, and potential effectiveness studies“, Journal of Medical Internet Research, vol. 25, p. e43293, 2023.
- Y. Jung, “Development and performance evaluation of chatbot system for office automation“, Asia-Pacific Journal of Convergent Research Interchange, vol. 8, no. 10, p. 37-47, 2022.
- J. Krueger and L. Osler, “Communing with the dead online: chatbots, grief, and continuing bonds“, Journal of Consciousness Studies, vol. 29, no. 9, p. 222-252, 2022.
- Z. Zeng, R. Song, P. Lin, & T. Sakai, “Attitude detection for one-round conversation: jointly extracting target-polarity pairs“, Journal of Information Processing, vol. 27, no. 0, p. 742-751, 2019.
- S. Jang, J. Kim, S. Kim, J. Hong, K. Kim, & E. Kim, “Mobile app-based chatbot to deliver cognitive behavioral therapy and psychoeducation for adults with attention deficit: a development and feasibility/usability study“, International Journal of Medical Informatics, vol. 150, p. 104440, 2021.
- S. Zhang, E. Dinan, J. Urbanek, A. Szlam, D. Kiela, & J. Weston, “Personalizing dialogue agents: i have a dog, do you have pets too?“, Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2018.
- M. Ng, J. Firth, M. Minen, & J. Torous, “User engagement in mental health apps: a review of measurement, reporting, and validity“, Psychiatric Services, vol. 70, no. 7, p. 538-544, 2019.
- Lakhani, A. (2023, July 14). Enhancing Customer Service with ChatGPT Transforming the Way Businesses Interact with Customers.
- Gupta, M., Akiri, C., Aryal, K., Parker, E., & Praharaj, L. (2023). From ChatGPT to ThreatGPT: Impact of Generative AI in Cybersecurity and Privacy. IEEE Access, 11, 80218–80245.
- How does ChatGPT work? | Zapier. (n.d.). How Does ChatGPT Work? | Zapier.
- Lalwani, T., Bhalotia, S., Pal, A., Rathod, V., & Bisen, S. (2018). Implementation of a Chatbot System using AI and NLP. SSRN Electronic Journal.
- Feine, J., Morana, S., & Maedche, A. (2020, September 6). A chatbot response generation system. Proceedings of the Conference on Mensch Und Computer.
- ChatGPT creates mutating malware that evades detection by EDR. (n.d.). CSO Online.
- C. Castro, “A discussion about the impact of chatgpt in education: benefits and concerns“, Journal of Business Theory and Practice, vol. 11, no. 2, p. p28, 2023.
- M. Zimmer, ““but the data is already public”: on the ethics of research in facebook“, Ethics and Information Technology, vol. 12, no. 4, p. 313-325, 2010.
- Yadav, K., Gupta, B. B., Chui, K. T., & Psannis, K. (2020). Differential privacy approach to solve gradient leakage attack in a federated machine learning environment. In Computational Data and Social Networks: 9th International Conference, CSoNet 2020, Dallas, TX, USA, December 11–13, 2020, Proceedings 9 (pp. 378-385). Springer International Publishing.
- Xu, Z., He, D., Vijayakumar, P., Gupta, B., & Shen, J. (2021). Certificateless public auditing scheme with data privacy and dynamics in group user model of cloud-assisted medical wsns. IEEE Journal of Biomedical and Health Informatics.
- Casillo, M., Colace, F., Gupta, B. B., Santaniello, D., & Valentino, C. (2021). Fake news detection using LDA topic modelling and K-nearest neighbor classifier. In Computational Data and Social Networks: 10th International Conference, CSoNet 2021, Virtual Event, November 15–17, 2021, Proceedings 10 (pp. 330-339). Springer International Publishing. https://link.springer.com/chapter/10.1007/978-3-030-91434-9_29
- Singla, A., Gupta, N., Aeron, P., Jain, A., Garg, R., Sharma, D., … & Arya, V. (2022). Building the Metaverse: Design Considerations, Socio-Technical Elements, and Future Research Directions of Metaverse. Journal of Global Information Management (JGIM), 31(2), 1-28.
Tiwari H. (2023) Chatbots and Data Privacy: Unfiltered Responses and Their Negative Impacts, Insights2Techinfo, pp.1