The advancement of the design of ALFRED beyond the conceptual phase, passes through the analysis of the impact of uncertainties, notably to what concerns safety-related conditions. Compliancy of plant safety to Design Extension Conditions is, according to IAEA and in line with the meaning itself of these beyond-design conditions, usually investigated by best estimates only. Due however to the demonstration nature of ALFRED, it was decided to assess the actual safety performances of this system even in ultimate conditions. To this regard, the emphasis was put on unprotected events like the UTOP (unprotected transient of over-power) and ULOOP (unprotected loss of offsite power, resulting from the combination of a loss of flow and loss of heat sink under unprotected conditions), pinpointed as the most challenging situations sought for the plant. The purpose of the present work, which has been divided in three parts, was then to assess the ultimate ALFRED safety margins against failure of the key core components and systems (Part III). To target this objective, the evaluation of uncertainties coming, on one hand, from nuclear data was performed at first, to retrieve their impact on the reactivity coefficients, thereby on the transient behavior driven by the latter (Part I); then, uncertainties from material properties, fabrication procedures, operation and computational tools were propagated to assess their influence on the thermal-hydraulics of the system (Part II). In this paper, presenting the first part of the work, the focus is on nuclear data. As such, a sensitivity/uncertainty analysis of the ALFRED core on key elementary reactivity effects, forming the basis for computing the feedback coefficients, has been performed. The sensitivity analysis allowed pointing out firstly the most relevant cross-sections for every response function. Uncertainty analysis allowed then establishing a possible range of confidence for the reactivity effects. The adjoint-based technique implemented in the TSUNAMI-3D module of the SCALE6 system was used. The confidence intervals identified for each reactivity effect have been combined then into confidence intervals for the feedback coefficients. Finally, the most conservative, yet physically sound (i.e., where correlations among coefficients, stemming from dependencies on common nuclear data, are taken into account), set of reactivity coefficients has been picked out from the confidence intervals, enabling transient calculations to propagate uncertainties into transient behavior. Using this off-nominal set, and comparing results with the reference ones, exploiting the system codes SIM-LFR and RELAP, a negligible effect of nuclear data uncertainties has been found for the ULOOP, while an increase of the maximum achieved power of around 6% has been computed for the UTOP. Overall, a modest contribution of nuclear data uncertainties for these transients has been found which, however, must be combined with the thermal-hydraulics one so to finally assess safety margins.