Whakaari/White Island’s sudden eruption in December 2019, which killed 22 people and severely injured 25 others, was New Zealand’s deadliest volcanic disaster in recent history.But it was not a freak event. Over the past 20 years, New Zealand has experienced another half-dozen sudden volcanic explosions with the potential to kill or injure people nearby.In many respects, these were near-misses, occurring at night or when few tourists or workers were present. Yet none was successfully forecast early enough to warn people or evacuate them beforehand.The danger posed by such sudden eruptions is not unique to New Zealand. In 2014, Japan’s Mt Ontake erupted with little warning, killing 63 people, many of whom had been hiking near the summit.Our new research suggests machine-learning techniques could help provide more warning. By detecting subtle changes in the continuous vibrations around a volcano, they can potentially identify signs of an impending eruption hours beforehand.But what if faster warnings also mean more false alarms? How many should we accept if earlier warnings can save lives?How many false alarms are too many?Under conventional warning systems, experts must interpret complex unrest signals and assess the level of risk before sounding the alarm.This is essential for understanding a developing volcanic crisis. But it faces a fundamental challenge when escalation occurs over minutes or hours rather than days or weeks.Automated warnings, drawing directly on real-time monitoring data, could complement expert judgement by responding rapidly when volcanic unrest suddenly changes.Volcano observatories have good reasons for caution. An algorithm might miss an eruption, detect signals scientists cannot yet fully explain, lack corroborating evidence or generate too many false alarms.Our new study asks how many false alarms might be worth accepting in exchange for a greater chance of warning people before an eruption.We re-analysed the performance of a machine-learning eruption forecaster at five volcanoes in New Zealand, Japan, Chile and Russia, using years of seismic data to estimate the probability of an eruption within a rolling 48-hour window.At Whakaari, when tested on data it had not seen during training, the forecaster anticipated four of five eruptions. The tradeoff was around 15 days each year when a warning would have been in place without an eruption occurring.Fifteen days of false warnings sounds substantial. But whether that is too many depends on what is at stake.We used a cost-loss model to compare the economic disruption caused by precautionary action with the losses that could potentially be avoided when an eruption was successfully forecast.The results suggest precautionary action could reduce preventable losses by 30–90%.The balance will differ between volcanoes. Closing Mt Ruapehu during the busy ski season, for example, could come at a steep price in lost revenue and disruption.But those costs need to be weighed against the deaths, serious injuries and lifelong consequences a timely evacuation could prevent.In some situations, a warning system that “cries wolf” more often may ultimately be safer – and economically rational – than one that waits for high confidence.The case for faster warningsOther warning systems already make similar trade-offs. Tsunami warnings, for instance, produce far more alarms than damaging waves, yet communities continue to evacuate when sirens sound.Good risk communication and public trust allow these systems to remain effective despite imperfect forecasts. Volcano monitoring may need to consider a similar tolerance for uncertainty, particularly where eruptions develop too quickly for conventional warning processes.Importantly, introducing automated warnings would not mean replacing volcanologists.The systems we propose depend on monitoring data collected by organisations such as New Zealand’s GeoNet. Human expertise remains essential for interpreting volcanic behaviour and managing a developing crisis.Automation may be most useful during sudden escalation, providing an initial warning while experts assess what is happening and determine what comes next.Earlier warnings could also help protect skifields, walking tracks and nearby communities, as well as roads, power lines and other critical infrastructure.Our modelling is not a comprehensive cost-benefit analysis of operational volcano monitoring. But it does show how the costs of false alarms can be weighed against the potential benefits of earlier action.This aligns with the United Nations’ Early Warnings for All initiative, which aims to provide universal access to multi-hazard warning systems by 2027.For volcanoes capable of erupting with little warning, waiting for certainty carries its own risk.Our results suggest accepting more false alarms may sometimes be a reasonable price for giving people more time to get out of harm’s way.Alberto Ardid has received research funding from the Ministry for Business, Innovation and Employment.David Dempsey has received research funding from the Ministry for Business, Innovation and Employment.Shane Cronin receives funding from the New Zealand Ministry of Business, Innovation and Enterprise. He is a member of the New Zealand Volcanic Science Advisory Group to the National Emergency Management Agency.Garry McDonald does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.