The Vera C. Rubin Observatory in Chile, whose ten-year Legacy Survey of Space and Time (LSST) began science operations in 2025, images the entire visible southern sky every few nights. Each time it detects something that changed - a star that brightened, a moving asteroid, a supernova, a flaring black hole - it issues an “alert.” Rubin generates millions of these alerts every single night, far more than any team of astronomers could ever inspect by hand.
To make that torrent usable, Rubin relies on a network of independent software systems called community brokers. Brokers ingest the full public alert stream and use machine learning to filter, enrich, and classify each event - distinguishing, for example, a supernova from an asteroid from an instrumental artifact, and attaching contextual data and a probable type. Researchers then subscribe to the slice of classified alerts relevant to their science and decide what to follow up with other telescopes. Established brokers such as ALeRCE, Fink, Lasair, ANTARES, and others were built and tested on earlier surveys like the Zwicky Transient Facility before Rubin came online.
Rubin is a defining case of AI as essential infrastructure for science: the alert volume is simply too large for the classification to be anything but automated, so machine learning sits on the critical path between the telescope and discovery.