Machine Learning Finds Ten Thousand Hidden Exoplanets
Staring at a night sky filled with stars, an observer might count a few thousand points of light. Being told that hidden among them, invisible to the naked eye, are ten thousand more worlds that were simply overlooked would be startling. That is precisely the situation astronomers found themselves in when they reexamined data from a satellite that had been scanning the heavens for years.
The instrument in question is NASA’s Transiting Exoplanet Survey Satellite, known simply as TESS. It watches stars for telltale dips in brightness - moments when a planet passes in front of its host star, briefly blocking some of the light. Since its launch, TESS alone has been credited with discovering nearly 900 confirmed exoplanets. [1] That number seemed impressive. It was, until researchers decided to look again with fresh eyes.
The old approach had a blind spot. TESS gathered data on millions of stars, but scientists typically focused on the brightest ones when searching for planets. Faint stars received less attention. They were harder to analyze, and the signals from any orbiting planets were weaker and more easily lost in noise. So those dimmer suns went largely unexamined, their potential planetary families hidden in plain sight within the data.
What changed was the method. A team of researchers applied machine learning - a form of artificial intelligence where computer algorithms learn to recognize patterns from examples - to comb through data from TESS’s first year of operations back in 2018. The AI was trained to spot the subtle signatures of transiting planets. When it finished its sweep, it had identified 10,091 possible planet-like objects that had never been observed before. Ten thousand new candidate worlds, waiting for confirmation.
To put that number in perspective, humanity has confirmed just over 6,000 exoplanets since the first one was discovered in 1995. The total confirmed count stands at 6,286, according to NASA’s Exoplanet Science Institute, which monitors and tracks these distant worlds. [2] Nearly 8,000 additional candidates were already awaiting official confirmation. If the new find is verified, it would more than double the number of known exoplanets in a single stroke.
The discovery was published on ArXiv, a research repository where scientists share their work before formal peer review. [1] That means the findings have not yet been independently verified by other researchers. But the potential implications are enormous. Finding ten thousand new worlds would represent a major leap forward in the hunt for planets beyond Earth that could support life.
The technique itself marks a shift in how astronomy is done. Instead of relying solely on human analysis of data, scientists are increasingly turning to artificial intelligence to find what human eyes might miss. The machine learning approach allowed the team to expand their search to include another 83 million fainter stars that TESS had already observed. That is a staggering number of additional targets, each one a potential home to planets that could harbor conditions suitable for life.
When Counting Every Star Becomes Impossible
The challenge astronomers face is one of scale. The Milky Way galaxy alone is thought to contain billions of planets. Even with dedicated space telescopes like TESS scanning the sky, the sheer volume of data overwhelms traditional methods of analysis. A human researcher can examine only so many light curves - the graphs that show how a star’s brightness changes over time - before fatigue sets in and subtle signals get missed.
This is where machine learning proves its worth. The algorithms do not tire. They do not get distracted. They process millions of data points with consistent attention, flagging anything that looks like the signature of a transiting planet. The team behind this discovery trained their AI to recognize patterns that might indicate a planet passing in front of its star, then set it loose on the vast archive of TESS data from 2018.
The results suggest that the previous census of exoplanets was incomplete in ways not fully appreciated. By restricting searches to bright stars, astronomers had been sampling only a fraction of what was actually out there. The faint stars, it turns out, may be just as likely to host planets. The discovery of 10,091 new candidates implies that our counting methods needed revision - that the models we used to estimate the prevalence of planets across the galaxy were based on incomplete data.
This is not the first time such a revision has been necessary. When the first exoplanet was confirmed in 1995, it upended long-held assumptions about how planetary systems form. That first discovery, made around a star called 51 Pegasi, showed that giant planets could exist in orbits much closer to their stars than anything in our own solar system. The discovery forced astronomers to rethink their models of planetary formation and migration.
The current find suggests another such moment may be approaching. If ten thousand new candidates are confirmed, the statistics of where planets exist will shift dramatically. The models that describe how common planets are around different types of stars will need to account for the fact that faint stars - the most numerous kind in the galaxy - may harbor far more planets than previously estimated.
The Search Intensifies as New Tools Arrive
The timing of this discovery is significant. NASA is preparing to launch its Nancy Grace Roman Space Telescope, with a potential launch date in 2026. This advanced observatory is designed to discover thousands of new exoplanets primarily through a technique called gravitational microlensing. That method relies on the gravitational field of a foreground star bending and magnifying the light of a background star, revealing planets that might otherwise remain undetected.
Gravitational microlensing works differently from the transit method used by TESS. Instead of watching for dips in brightness, it watches for brief brightenings caused by gravitational lensing effects. This allows it to detect planets that are farther from their host stars, planets that transit observations might miss entirely. The Roman telescope could unlock even more new worlds, complementing the discoveries made by TESS and expanding our understanding of the variety of planetary systems in the galaxy.
The search for exoplanets is not merely an exercise in counting. It is driven by a fundamental question: are we alone in the universe? Since the first exoplanet was discovered three decades ago, researchers have spent years looking for worlds that might harbor life. They have found strange and captivating planets - gas giants with bizarre orbits, rocky worlds scorched by their stars, and ice balls drifting in the darkness.
But there is one thing they have not yet found: a planet that closely resembles Earth. Such a landmark discovery would be a monumental leap forward in the search for a world that could potentially support living organisms. In April 2025, scientists came close. An exoplanet known as K2-18b achieved a degree of fame when a team of astronomers claimed to have found in its atmosphere what they called the strongest evidence yet that life exists anywhere else besides Earth.
That claim has since been cast into doubt. Other scientists have questioned the findings, putting a damper on the notion that humanity finally had proof we are not alone in the cosmos. The episode illustrates both the promise and the peril of exoplanet research - each discovery brings hope, but verification is a slow and painstaking process. The ten thousand new candidates from the TESS data will face the same scrutiny before any of them can be confirmed as genuine planets.
The search for life beyond Earth is also taking place much closer to home. While Mars is considered a potential time capsule with evidence of ancient life, other planets in our solar system are being targeted for exploration. Jupiter and Saturn are orbited by moons that NASA spacecraft are investigating in the hunt for life. The Jovian moon Europa, where a probe is due to arrive in 2030, is a prime target - scientists want to search beneath its icy surface for signs of water and habitability.
Saturn’s moon Enceladus, which measures about 310 miles wide, is also thought to be suitable for life. These moons, with their subsurface oceans and potential hydrothermal activity, offer a different path in the search for living organisms. They remind researchers that the question of life beyond Earth is not limited to distant exoplanets - it extends to the frozen worlds orbiting our own sun. The discovery of ten thousand new candidate planets adds urgency to these efforts, expanding the list of places where life might one day be found.
Sources
1. NASA