A Comprehensive Analysis of Risk and Trust in Using GPT-Style Models for E-Learning Systems

The incorporation of GPT-style LLMs (Large Language Models) into e-learning mechanisms has changed digital means of teaching by allowing the generation of content in a personalised manner, smart tutoring, and automated evaluation. But despite this, concerns regarding trust as well as risk, data privacy, bias, reliability of content, and dependence of the learners are continuing to be observed from a methodical and experiential point of view, which is often insufficient. This study offers a detailed exploration of trust-risk interaction in connection with adopting GPT-style frameworks in e-learning settings. A hybrid research mechanism is implemented, bringing together an extensive survey conducted on 412 instructors as well as students through experimental evaluation in three main e-learning approaches, which are the generation of content, adaptive tutoring, and acquiring feedback through automation. Statistical evaluation, in addition to checking the reliability, correlation analysis, and regression modeling, is utilised to calculate trust and risk indicators. This survey denotes robust underlying consistency with a value of 0.87 Cronbach's alpha. The results acquired show that the subjective adequacy with β = 0.42 and p < 0.01, as well as explainability with β = 0.31 and p < 0.05, impact the reliability of consumers in a positive manner, whilst risk of data privacy with β = − 0.38 and p < 0.01, and content generated via AI significantly reduce the levels of trust. Preliminary findings denote that learning via the assistance of GPT has increased the percentage of task completion activities by 23%, but despite that, there was a reduction in accuracy by 14% as spotted during evaluation in risky scenarios without human surveillance. On the basis of these evaluations, this paper presents a risk-trust mechanism that combines transparency frameworks, validation through HITL (Human-in-the-loop), and surveillance controls. The research summarizes that GPT-style mechanisms also provide trustworthy benefits in e-learning environments, fostering it in a sustainable manner requires reducing risks and fortifying trust in a systematic manner via supervision, technical, and moral protection.