Price instability is recognized as a major driver of food insecurity that may have negative impacts on the social, economic and health status of many households. Since the food price spikes of 2007–8, prices of the main food and feed crops have shown high variability in the international and local markets for almost a decade. Although agricultural commodity prices are influenced by many factors, shocks on agricultural productions are often recognized as a major driver of price volatility. Adverse environmental conditions and the resulting decline in regional production have be shown to have greatly contributed to spikes in global food prices. The connection between production levels, agricultural commodity prices and food security has led many companies, governments and international organizations to develop forecasting systems to anticipate production shortages or, on the contrary, overproduction. The last decade has seen a rising number of applications of machine learning, with a particular focus on anticipating agricultural production shocks. The reasons are multiple. First, machine learning techniques can now be more easily applied than ever before using specialized packages implemented in open source packages. Second, large agricultural databases are becoming increasingly easily accessible, thanks in large part to the emergence of good practices in open science. Finally, mechanistic models were found to miss several major yield losses in Europe and in the US which led scientists to look for alternative solutions. For all these reasons, different machine learning tools are now being used more often to analyse production shocks, identify their origins, and assess their consequences on food security. Based on a brief review of the literature, we identify the main biotic and abiotic factors determining crop production shocks, propose a hierarchy for the factors with the most impact, and show that these factors are often interdependent, making crop yield forecasting rather difficult. Next, based on our own recent research projects, we show how the use of machine learning and of probabilistic models could open – in connection with the development of open access databases and new powerful algorithms – a new avenue for predicting production shocks and their impacts on agricultural prices. Lastly, we discuss the economic mechanisms that link together seasonal production forecasts and market impacts and how improving production forecasts could reduce price volatility and the risk of abrupt shifts in price trends.