Modern power systems are increasingly vulnerable to stealthy false data injection attacks (FDIAs) targeting state estimation. Executed without the operator’s knowledge, these attacks can lead to severe consequences, including economic losses, power outages, and full system blackouts. Previously, it was believed that executing destructive FDIAs requires full understanding of the operator’s response to such attacks, which demands in-depth knowledge of power system details, including generation prices and system constraints. Nevertheless, this paper reveals a noteworthy vulnerability wherein an attacker can employ publicly accessible historical locational marginal price (LMP) data and either all or a limited subset of phasor measurement unit (PMU) data to construct a predictive model. This model, utilizing a neural network, takes the loading profile of a section or the entire system as its input to forecast the LMPs. Such a model can be applied in designing high-damage FDIAs. Rigorous testing on the IEEE 30-, IEEE 57-, and IEEE 118- bus test systems demonstrates the model’s ability to predict the LMPs of the system with impressive accuracy and precision. This finding not only validates the model’s efficacy but also underscores the pressing need to address the newfound vulnerability of modern power systems against attackers utilizing public data.