A team of astronomers at the University of Warwick has pulled 118 confirmed exoplanets out of four years of archived NASA satellite data using an automated AI pipeline, and 31 of those worlds had never been flagged by any previous search. The tool, called RAVEN (RAnking and Validation of ExoplaNets), scanned full-frame images from NASA’s Transiting Exoplanet Survey Satellite (TESS), checking more than 2.2 million Sun-like stars for the tiny, repeating dips in brightness that betray an orbiting planet. The results, detailed in a preprint posted in June 2026, represent one of the largest single batches of validated exoplanets to come from TESS data and offer a fresh statistical picture of how common close-in planets really are.
To put the haul in perspective: TESS has confirmed roughly 500 exoplanets since its 2018 launch. Adding 118 in one sweep, with 31 of them entirely new, is the kind of acceleration that only machine-driven analysis can deliver.
How RAVEN hunts for planets
RAVEN pairs two machine-learning components with a statistical framework built to tell real planetary signals apart from imposters. A gradient-boosted classifier and a Gaussian-process model score each candidate signal, while a Bayesian layer calculates the probability that the signal is genuinely planetary rather than, say, an eclipsing binary star or an instrumental glitch. The system was trained on simulated transits and on a curated set of confirmed false positives drawn from real TESS observations, giving it a double check against the kinds of noise that have historically bogged down human reviewers.
The search covered TESS full-frame image sectors 1 through 55, spanning observations from mid-2018 through mid-2022. The team focused on orbital periods between 0.5 and 16 days across a magnitude-limited sample of stars. That short-period window captures the planets most likely to produce detectable transits: hot Jupiters, super-Earths hugging their stars, and the rare worlds that sit inside the so-called Neptunian desert, a zone where Neptune-sized planets are surprisingly scarce because intense stellar radiation is thought to strip their atmospheres away.
What the new planets tell us
Many of the validated worlds are roughly Neptune-sized or smaller, orbiting FGK main-sequence stars similar in mass and temperature to the Sun. Because they circle so close to their host stars, these planets are scorching hot, receiving many times the stellar radiation that Earth does. That rules them out as habitable real estate, but it makes them prime laboratories for studying how planetary atmospheres behave under extreme irradiation and how close-in planets form and migrate in the first place.
Some of the most scientifically valuable finds sit inside the Neptunian desert. Standard models of atmospheric escape predict that mid-sized planets orbiting this close should be stripped down to rocky cores or destroyed entirely. Every confirmed planet found surviving in that zone forces theorists to sharpen their models of planetary evolution. RAVEN’s ability to sift through millions of light curves at speed raises the odds of catching these rare outliers, which carry outsized scientific weight.
A companion demographics study used the RAVEN-validated sample to estimate how frequently close-in planets orbit Sun-like stars. That analysis produced an overall occurrence rate of roughly 9.4 percent for planets with orbital periods under 16 days, a figure that tightens earlier estimates and feeds directly into target lists for upcoming missions such as ESA’s PLATO spacecraft, scheduled for launch in 2026.
Where the numbers need scrutiny
The headline figures carry a wrinkle worth understanding. The primary preprint reports 118 newly validated planets. The University of Warwick’s press release frames the result as “over 100 exoplanets” validated, with 31 described as newly detected worlds. The most likely reading: 118 planets passed RAVEN’s statistical threshold, but only 31 were genuinely new detections rather than independent confirmations of candidates already flagged at lower confidence by other searches. No public table yet cross-matches the 31 new planets with their coordinates, orbital periods, and host-star properties in a way that fully resolves the gap between the two numbers.
That distinction matters. For researchers planning follow-up observations with the James Webb Space Telescope or ground-based radial-velocity instruments, knowing whether a planet is a fresh discovery or a firmed-up prior candidate determines how much precious telescope time it deserves.
The 9.4 percent occurrence rate also deserves careful framing. It comes from a demographics preprint built on the RAVEN candidate sample, and whether the rate holds when the pipeline is extended to TESS sectors 56 and beyond remains an open question. If a bump in the shortest-period bin (0.5 to 4 days) persists in later data, it could imply that the boundary of the Neptunian desert is sharper than current models predict. That hypothesis has not yet been tested against newer sectors.
There is also the inherent limitation of any machine-learning classifier: it can only be as good as its training data. Rare configurations, such as triple-star systems or background eclipsing binaries blended with bright foreground stars, may still fool the filters or cause genuine planets to be thrown out. The RAVEN authors acknowledge this, and independent checks by other teams will be essential before the validated planets earn formal catalog designations.
What comes next for automated planet hunting
All three papers anchoring the RAVEN results are preprints hosted on arXiv, meaning they have been made publicly available but have not yet completed formal peer review. Preprints are standard practice in astrophysics and often contain reliable results, but they can still be revised during the review process. Based on citation trails already visible, the peer-reviewed versions will likely appear in Monthly Notices of the Royal Astronomical Society.
RAVEN is not an official NASA tool. It is an independent academic product, and its results will need to pass through the standard community vetting process, including independent replication and, where feasible, radial-velocity mass measurements, before the planets are formally cataloged. NASA’s TESS continues to collect data in extended mission phases, which means the same automated methods can be reapplied to newer sectors without waiting for labor-intensive manual review.
The broader shift is hard to overstate. TESS alone has generated hundreds of thousands of candidate signals, far more than any group of human vetters can reasonably inspect. Tools like RAVEN act as triage systems, promoting the most promising candidates while filtering out stellar variability, instrumental artifacts, and statistical noise. Future missions designed to find Earth-sized planets in wider orbits, along with massive ground-based surveys, will lean even more heavily on this kind of probabilistic validation. The RAVEN results are an early, concrete demonstration that automated pipelines can deliver statistically robust planet samples at scale. How many of these 118 worlds survive peer review and follow-up scrutiny will be the real test, and the answer should start arriving within the year.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
