NASA’s Hubble Space Telescope continues to reveal the secrets of the universe, three and a half decades after its launch, as a new AI technique from the European Space Agency (ESA) identifies more than 800 previously unreported space objects.
Launched in 1990 and still in operation, the Hubble Space Telescope has amassed a vast archive of data over its decades spent scanning the universe. Now, a recent paper in Astronomy & Astrophysics reports the identification of 1,300 unusual objects by AI in just two and a half days, many of which have never appeared in the astronomical literature.
Hubble Image Analysis
The astronomers behind the new research fed their AI neural network 100 million image cutouts from the Hubble Legacy Archive. These cutouts are tiny—only dozens of pixels wide—yet the AI was able to meaningfully determine what they contained.
Galaxies accounted for most of the anomalous cutouts, typically undergoing mergers or other unusual interactions that distorted their morphologies or left them trailing long streams of stars and gas. Other cutouts were identified as spacetime distortions that bent light from distant galaxies into arcs or rings before it reached Hubble.
More unusual cosmic features were also identified, including galaxies resembling jellyfish with strange tentacle-like streams of gas, and edge-on planet-forming disks that appear like hamburgers in the images (several of which can be seen above).
Perhaps most notably, dozens of the objects were so bizarre that they did not fit any existing classification.
Developing AnomalyMatch
The sheer scale of data amassed in the continually expanding archives of Hubble and other telescopes has become so large that it defies traditional human-led analysis. These advanced space observatories generate data at a rate far exceeding that of earlier eras in astronomy. Prior attempts to open the data to citizen scientists have met with some success, but the backlog has grown too large for those initiatives to fully address. Until now, astronomers have relied on manual analysis and the occasional fortunate discovery made while examining specific targets.
While Hubble was constructed by NASA, ESA researchers David O’Ryan and Pablo Gómez developed the solution leveraged by the team for its analysis of this torrent of data. Specifically, the pair created a neural network called AnomalyMatch, which is designed to examine collections of imagery far more quickly than humans can. Still, AnomalyMatch was trained to perform pattern recognition similar to that of humans, allowing it to “learn” to identify rare and unusual objects hidden in the data.
“Archival observations from the Hubble Space Telescope now span 35 years, offering a rich dataset in which astrophysical anomalies may be hidden,” said lead author David O’Ryan.
Exploring Hubble and Other Telescopes
Although the team’s initial paper focuses on data retrieved using the Hubble Space Telescope, the technique can be applied more broadly as well. Other platforms, such as ESA’s Euclid mission, are generating vast amounts of data that AnomalyMatch could help process. Crucially, O’Ryan and Gómez’s neural network results have been manually reviewed by astronomers, who have so far confirmed 1,300 of the anomalies flagged by AnomalyMatch.
“This is a powerful demonstration of how AI can enhance the scientific return of archival datasets,” co-author Pablo Gómez said. “The discovery of so many previously undocumented anomalies in Hubble data underscores the tool’s potential for future surveys.”
Future advanced space observatories will also benefit from the implementation of AnomalyMatch. For instance, NASA’s upcoming Nancy Grace Roman Space Telescope will provide a wider field of view than the James Webb Space Telescope, offering yet another enormous data set for astronomers to search through in the years ahead.
The paper, “Identifying Astrophysical Anomalies in 99.6 Million Source Cutouts from the Hubble Legacy Archive using AnomalyMatch,” appeared in Astronomy & Astrophysics on December 16, 2025.
Ryan Whalen covers science and technology for The Debrief. He holds an MA in History and a Master of Library and Information Science with a certificate in Data Science. He can be contacted at ryan@thedebrief.org, and follow him on Twitter @mdntwvlf.