Artificial intelligence has just turned one of astronomy’s most familiar workhorses into a discovery engine all over again. By trawling through decades of archival images from the Hubble Space Telescope, a neural network has flagged more than 1,300 strange objects that do not fit neatly into existing catalogs or classifications. The haul ranges from colliding galaxies to warped rings of light, and it hints at how much of the universe has been hiding in plain sight.
The project effectively gives Hubble a second life, not by launching new hardware but by reinterpreting what it has already seen. Thirty to 35 years of observations, once considered “old” data, are suddenly a frontier for fresh science as AI tools learn to spot the rare, the odd and the previously overlooked.
How AI turned dusty archives into a discovery goldmine
The key to this reawakening is a neural network called AnomalyMatch, designed to sift through the Hubble Legacy Archive and highlight anything that looks unusual compared with the telescope’s vast training set of “normal” galaxies and stars. Instead of asking the algorithm to find a specific type of object, astronomers asked it to surface outliers, the digital equivalent of circling every weird smudge on a photographic plate. In practical terms, that meant feeding the system images spanning the Hubble Space Telescope’s long lifetime and letting it rank the most statistically surprising features for human review, a process that would have been impossible to do by eye at this scale.
Earlier this year, researchers reported that the scan had uncovered more than 1,000 such anomalies in the Hubble Space Telescope’s image archive, with some objects so odd that they defied existing classification schemes. A separate analysis framed the scale of the effort in time, noting that Thirty five Years of Data from Hubble had been combed to reveal 1,300 Anomalies, underscoring how much latent discovery power was locked in the archive.
What the “weird” objects actually are
Once the algorithm had done its triage, astronomers began sorting the anomalies into physical categories, and a pattern quickly emerged. Around 50 percent of the flagged objects turned out to be galaxies in the act of merging, their disks stretched and twisted by gravity as they collide. Others showed the telltale arcs and rings produced by gravitational lensing, where the mass of a foreground galaxy bends the light of a more distant one into distorted shapes that can look like cosmic parentheses. These are exactly the kinds of rare, high value systems that cosmologists use to probe dark matter and galaxy evolution, and they are easy to miss when buried among millions of more ordinary spirals and ellipticals.
Some of the strangest finds have been nicknamed “jellyfish” galaxies, with long tendrils of gas and newborn stars streaming away from a compact core, likely sculpted by the pressure of hot plasma in galaxy clusters. A recent release from NASA highlighted a collection of six images of galaxies uncovered in this way, including systems where lensing turns a background galaxy into arcs or rings around a massive foreground object. Other anomalies include planet forming disks seen edge on and galaxies with warped or asymmetric structures that hint at past interactions.
The numbers behind the anomaly boom
Different teams tally the results slightly differently, but they all point in the same direction, toward a bonanza of oddities. One group describes “over 1,300” anomalies in roughly 30 years of Hubble data, a figure echoed in the reporting that first brought the scale of the search to public attention. Another analysis, focused on the full mission span, emphasizes that Hubble AI Scan roughly the same order of magnitude of anomalies across 35 years of observations. A separate count from European researchers speaks of 1400 quirky objects in Hubble’s archive, a figure that reflects both the AI’s output and subsequent human vetting.
Within that broader total, one focused search turned up 800 never before seen “cosmic anomalies” in old Hubble images, including Six previously undiscovered astrophysical objects that had never been described before. Another team, working with a slightly different selection of fields, reported that their AI assisted method had attracted 780 views and 46 likes on an outreach page summarizing their datasets, a reminder that public fascination with these oddities is part of the story too.
Inside AnomalyMatch, the neural network behind the haul
At the heart of this effort is a specific implementation of machine learning that treats anomaly hunting as a pattern recognition problem. Neural networks, a form of artificial intelligence, are trained on labeled examples of galaxies, stars and other objects until they can reconstruct what a “typical” Hubble scene looks like. Anything that deviates strongly from that learned template is flagged as a candidate anomaly. The Space Telescope Science Institute has described how these Neural networks can be tuned to search for rare galaxy mergers and jellyfish like structures, effectively turning the Hubble Space Telescope into a targeted survey instrument long after the images were taken.
The work by Ryan and colleagues, who applied AnomalyMatch to the Hubble Legacy Archive, shows how this approach scales. One account notes that AnomalyMatch completed its review in two and a half days and presented promising candidates for anomalies, with the system’s performance detailed in a report that explains how it finished the scan and what happened After human astronomers took a closer look. Another overview of The AI Revolution in Sky Scanning notes that Researchers David O’Ryan and Gómez used the tool to uncover everything from colliding galaxies to unclassifiable oddities in Hubble Space Telescope.
Why anomaly hunting matters for the future of astronomy
For astronomers, the payoff from this anomaly hunt is not just a bigger catalog, it is a sharper sense of what is possible in the universe. Some of the newly flagged systems will become laboratories for testing theories of dark matter, galaxy evolution and star formation under extreme conditions. Others may represent entirely new classes of objects, the kind of surprises that historically have driven major shifts in astrophysics. One overview of the project urges readers to Think of weirdly shaped galaxies, light warped by massive objects and planet forming disks seen edge on, all now surfaced systematically rather than by chance.
The approach is already influencing how future surveys are planned. As larger facilities come online, from the Vera C. Rubin Observatory to next generation space telescopes, the volume of data will dwarf Hubble’s archive, making manual inspection impossible. The success of AI tools on Hubble shows that anomaly detection can be baked into the pipeline from the start, turning every survey into a search for the unexpected. In that sense, the hundreds of anomalies uncovered so far are a proof of concept for a broader shift, one that treats AI not as a replacement for human curiosity but as a way to aim it more precisely at the universe’s strangest corners.
More from Morning Overview