TIRESIAS can provide much more precise predictive information compared to previous work, allowing companies to deploy truly proactive countermeasures based on the predicted information. For example, this system could predict which particular CVE will be used by an attacker when mounting a multi-step attack against a server, or assess the potential severity of an attack by only looking at its early steps. We have tested TIRESIAS on a dataset of 3.4 billion security events collected by a commercial intrusion prevention system from a daily population of 740k machines over a period of 27 days. We show that TIRESIAS is effective in predicting the specific malicious event that will occur with a precision of up to 0.93. We have seen that the models trained by TIRESIAS are fairly stable, with the detection performance changes slowly.