Large language models (LLMs) are transforming the way we work, learn, and access information. As our dependence on these tools grows, it becomes crucial to enhance their accuracy and ensure they align with our ethical standards. The most high-performing language models are currently trained and refined with the help of explicit human feedback. Here we propose a study that investigates the feasibility of implicit human feedback through passive brain-computer interfaces (pBCIs). Two calibration paradigms for moral judgment and error-perception elicitation and detection are described. The obtained classification models will be tested in an application phase with simulated chatbot conversations. If proven successful, pBCIs could provide novel and informative human implicit feedback in the process of LLM development.