The Vera Rubin Observatory:
How AI Is Making the World's Most Ambitious
Telescope Actually Work

The Machine Itself:
What Makes Rubin Unprecedented

What makes Rubin extraordinary is not just its size — it is the combination of aperture, field of view, and cadence that no previous telescope has achieved. Large telescopes typically have narrow fields of view; wide-field survey telescopes typically have small apertures. Rubin has both. Its etendue — the technical measure of a telescope's ability to collect light from a large area — is 319 m²·deg², compared to roughly 9 m²·deg² for the previous state-of-the-art Zwicky Transient Facility. It is not an incremental improvement. It is a categorical leap.

The first images, released publicly on June 23, 2025 — at watch parties on six continents, attended by people from 28 countries — revealed thousands of objects in single frames that had never been catalogued. Over 2,000 new asteroids appeared in the first release alone. One of them, 2025 MN45, turned out to be unusually large and rotating at a remarkable rate. The first peer-reviewed paper based on LSST Camera data identified an asteroid nearly the size of eight football fields completing a full rotation every two minutes — one of the fastest-rotating large objects in the Solar System ever observed.

The Data Problem:
Seven Million Signals a Night

Every night, Rubin will generate approximately 20 terabytes of raw images. Those images feed into a pipeline that compares each new frame with a template — a deep, combined reference image of the same patch of sky built from previous observations. The template is digitally subtracted from the new image. What remains, after subtraction, is only change: a star that brightened, a galaxy that dimmed, an object that moved. Each change becomes an alert.

At full operation, this process will generate up to 10 million alerts per night, each one delivered within two minutes of the image being captured. The system issued its first 800,000 alerts on the night of February 24, 2026 — the beginning of what the observatory calls its "real-time discovery machine." The alerts stream to an ecosystem of specialised software systems called brokers, developed by institutions worldwide, whose job is to classify, prioritise, and route the alerts to the scientists who need them.

The alert pipeline itself was developed over a decade by the University of Washington's Institute for Data Intensive Research in Astrophysics and Cosmology (DiRAC), a team of roughly two dozen researchers and software engineers. The pipeline processes the raw image stream, performs image differencing, detects change signals above noise thresholds, characterises each detection, and packages it as an alert — all within minutes of observation. It is one of the largest and most complex real-time data processing systems in the history of astronomy.

What Rubin Is Built to Find:
Six Targets That Will Reshape Astronomy

SELDON, ALeRCE & the
Broker Wars

Building AI for Rubin is not a single project — it is a global competition and collaboration between teams that have been developing, testing, and iterating their systems on precursor surveys like the Zwicky Transient Facility for years in preparation for the LSST data stream.

The ALeRCE broker (Automatic Learning for the Rapid Classification of Events), developed in Chile, has been operating on Zwicky Transient Facility data since 2019. Its machine learning classifiers annotate every Rubin alert with probabilities across multiple object classes — and it was one of the first brokers to classify alerts in Rubin's inaugural February 2026 alert stream. Lasair, the UK broker based at the University of Edinburgh, provides a complementary annotation stream with particular strength in supernova classification and cross-matching with external catalogs.

At UCL's Mullard Space Science Laboratory, researchers have spent years developing AI classification pipelines for Rubin specifically — training models on simulated supernova light curves to handle the millions of candidates the LSST will produce nightly. The challenge is not just classification accuracy but calibrated uncertainty: a broker that confidently misclassifies a genuinely novel object as a known type would cause scientists to miss it entirely. The systems must know not only what something probably is, but how confident that classification should be.

The First Discoveries:
What Rubin Found in Its Opening Weeks

Even before full survey operations begin, the early commissioning and First Look data have already produced genuine scientific results — a preview of what a decade of nightly observing will deliver.

What Rubin Can —
And Cannot — Tell Us

The most important caveat about Rubin is also the most exciting one. The observatory was designed to answer specific, well-defined questions — dark energy, dark matter, NEOs, supernovae. But the history of survey telescopes is overwhelmingly a history of unexpected discoveries. The Sloan Digital Sky Survey set out to map galaxies and ended up discovering quasars, stellar streams, exoplanet transits, and gravitational lenses that no one had predicted it would find.

Rubin will be orders of magnitude more sensitive. Its AI systems are being designed with explicit anomaly detection capability — the ability to flag objects that match no existing classification model, not to dismiss them as noise, but to flag them precisely because they are unusual. The most scientifically significant discoveries Rubin makes over its decade-long survey may be things that do not yet have a name.

The Decade Ahead:
Five Things to Watch

• 01

  The first interstellar object caught in time

  'Oumuamua passed through the inner Solar System in 2017 before astronomers could mount a serious follow-up campaign. Borisov in 2019 gave more warning but was already on its outward trajectory. Rubin's alert system — issuing detections within two minutes of observation — is the first survey fast enough and deep enough to catch an interstellar visitor early. When it happens, the scientific and cultural significance will be enormous.

• 02

  The dark energy verdict

  Recent data from DESI and JWST has suggested that dark energy may not be the cosmological constant Einstein proposed — it may be changing over time. Rubin's 10-year Type Ia supernova dataset, combined with its weak lensing measurements, will be one of the most powerful tests of this hypothesis ever assembled. The result will either confirm the standard model of cosmology or force a fundamental revision.

• 03

  Planetary defence at scale

  NASA's mandate under the Planetary Defense Act is to catalogue 90% of near-Earth objects larger than 140 metres. Current surveys are behind schedule. Rubin's expected 36,500 new NEO discoveries over 10 years, combined with real-time hazard assessment AI, represents the single largest contribution to planetary defence ever made. The question of whether that is enough — and how quickly warnings translate into action — is political as much as scientific.

• 04

  AI anomaly discovery — finding what we didn't know to look for

  The next five years will see broker systems increasingly move beyond classification of known object types toward genuine anomaly detection — flagging events that do not fit any existing model. The Rubin community is actively building these systems. The first true discovery from an AI anomaly flag — an object that would have been missed by any human-defined search — will mark a qualitative shift in how astronomy is done.

• 05

  The data legacy beyond Rubin

  Rubin's 500-petabyte dataset will outlive the survey itself by decades. The AI tools developed to process it — transient classifiers, galaxy morphology models, time-series foundation models — will transfer to the next generation of surveys: the Extremely Large Telescope, the Square Kilometre Array, the Nancy Roman Space Telescope. Rubin is not just a scientific mission; it is a training ground for the AI infrastructure that will underpin astronomy for the rest of the century.