The relatively high production costs of Easy Language (EL) texts in terms of time, financial and human resources, together with the nonexistence of EL guidelines for certain languages, or simply lack of awareness about them, may lead institutions to leave the needs of EL users unattended. In prior studies, authors have put forward Neural Machine Translation (NMT) as a tool that could bridge the resource and knowledge gap in such cases. Nevertheless, not only does raw NMT output produced with generic systems like DeepL present translation quality issues, but it also violates some of the most popular EL guidelines (Kaplan et al. 2019; Kaplan 2021). This seems to be confirmed by Rodríguez Vázquez et al. (forthcoming), who conducted a comparative study about Easy Language NMT across different text domains and languages. In this paper, data from the aforementioned study will be explored more in detail to answer the following questions: what is the relationship between translation issues and EL rule violations? Are translation errors derived from the application of EL best practices in the source text? Are specific EL non-compliance issues caused by translation errors produced by the system? Could the latter be simply language variations (e.g. transformation of active form into passive form), rather than translation errors? Additionally, we will draw upon the data to expand on the ongoing discussions around MT ethics (Nurminen and Koponen 2020) by putting EL target groups in the spotlight. For instance, the use of raw MT output is acceptable, even if it contains errors, for gisting purposes (Forcada et al. 2018). Would this be an option for all EL target populations? What would be the ethical implications? These questions will be addressed from MaaB and Rink (2020) understanding of EL as an empowering tool that allows both for participation and knowledge building.