NASA Says AI Found Hundreds Of Hidden Galaxies And Cosmic Oddities In Hubble Data

by Aaron Leong — Wednesday, January 28, 2026, 10:01 AM EDT

Deep in the digital bedrock of 35 years of Hubble observation archives, a new AI tool has uncovered more than 800 previously undocumented cosmic anomalies in a scant two and a half days.

Using a sophisticated neural network called AnomalyMatch, astronomers have unearthed nearly 1,400 bizarre astronomical objects hidden within the Hubble Space Telescope’s massive archives. Impressively, the AI sifted through nearly 100 million image cutouts from the Hubble Legacy Archive in less than three full days. The results, published in Astronomy & Astrophysics, represent the first time the entire dataset has been systematically searched for anomalies. Of the nearly 1,400 anomalies confirmed by researchers David O’Ryan and Pablo Gómez, more than 800 had never been documented in scientific literature.

esa ai — Six previously undiscovered astrophysical objects are displayed in this new image from NASA’s Hubble Space Telescope. They include three lenses with arcs distorted by gravity, one galactic merger, one ring galaxy, and one galaxy that defied classification. (Credit: NASA, ESA, David O'Ryan (ESA), Pablo Gómez (ESA), Mahdi Zamani (ESA/Hubble))

The mega-haul of cosmic oddities is as diverse as it is strange. Among the discoveries are rare jellyfish galaxies, which sport trailing tentacles of gas and stars as they plow through dense galaxy clusters, and ring galaxies formed by head-on galactic impacts. The AI also flagged dozens of gravitational lenses: massive foreground objects that act as cosmic magnifying glasses, warping the light of distant galaxies into surreal arcs and circles. Perhaps most intriguing to the team were several dozen objects that defied any known classification system entirely, which present a tantalizing puzzle for future study.

Tools like AnomalyMatch address a growing crisis in modern astronomy: the data deluge. As telescopes become more powerful, they generate far more information than human eyes can ever hope to manually inspect. O’Ryan and Gómez trained AnomalyMatch to mimic the human brain’s ability to recognize patterns, but obviously one that doesn't tire or get distracted. By narrowing down 100 million images to a manageable shortlist of 1,400 candidates, the AI acted as a high-speed filter, allowing the team to focus their expertise only on the highest-ranked anomalous candidates.

As we move into an era of even larger surveys, such as those expected from the James Webb and Euclid telescopes, tools like AnomalyMatch are already proving worthy assets. Indeed, they represent a fundamental shift in how we explore the stars: the future isn't just about building bigger lenses, but about developing sharper digital eyes and processing to find the needles in our ever-expanding cosmic haystack.