AI Uncovers 100+ Hidden Planets in NASA TESS Data, Including Rare Worlds

Astronomers at the University of Warwick have used an artificial intelligence tool called RAVEN to confirm more than 100 exoplanets hiding in NASA's TESS mission data - including 31 brand-new worlds and thousands of additional candidates in a region of space where planets were thought to be vanishingly rare.

The team trained RAVEN on observations from more than 2.2 million stars collected during the Transiting Exoplanet Survey Satellite's first four years of operation, specifically hunting for rare planets with orbital periods of less than 16 days. The results, reported via ScienceDaily, were published in the Monthly Notices of the Royal Astronomical Society.

How RAVEN Found What Humans Missed

TESS surveys huge swaths of the sky by watching for the tiny dips in starlight that occur when a planet crosses in front of its host star. The challenge is volume: with millions of stars under observation, separating real planetary transits from stellar noise, instrument artifacts, and look-alike signals like eclipsing binaries is a significant bottleneck.

RAVEN was designed to attack that bottleneck. According to the University of Warwick team, the AI worked through TESS observations of more than 2.2 million stars and confirmed more than 100 exoplanets, 31 of which had never been catalogued before. Crucially, the system also flagged thousands of additional candidates that will need follow-up observations before they can be promoted to confirmed status.

A Closer Look at the Neptunian Desert

One of the most intriguing aspects of the work is where many of these candidates sit. The so-called Neptunian desert is a region of parameter space - short orbital periods combined with Neptune-like sizes - where astronomers have historically found very few planets. The leading explanation is that intense stellar irradiation strips the atmospheres of Neptune-sized worlds that orbit too close to their stars.

By focusing the AI on planets with orbital periods under 16 days, the Warwick team was able to surface thousands of additional candidates within and around this desert. If even a fraction are confirmed by independent methods, they could refine - or complicate - current models of how short-period planets form, migrate, and lose their atmospheres.

Why This Matters Beyond the Headline Number

The headline figure - 100-plus confirmed planets - is impressive, but the methodological story may matter more in the long run. Machine learning vetting tools have been part of exoplanet science for years, but RAVEN's results suggest AI is now mature enough to do more than rank candidates: it can systematically recover real planets that conventional pipelines flagged as ambiguous.

That has practical consequences for upcoming missions. As ground- and space-based observatories continue to generate ever-larger datasets, similar AI-driven vetting workflows are likely to become standard rather than experimental. Each confirmation also tightens statistical estimates of how common short-period and Neptunian-desert planets actually are - a key input for understanding planetary system architectures across the galaxy.

What's Next

The thousands of additional candidates identified by RAVEN now need follow-up - typically a mix of higher-resolution imaging, radial velocity measurements, and analysis of additional TESS sectors - before they can be added to the confirmed planet tally. Expect a steady drip of follow-up papers rather than a single dramatic announcement.

For now, the Warwick result stands as both a planetary haul and a proof of concept: trained carefully and pointed at the right slice of the data, AI can pull rare worlds out of noise that humans, and earlier algorithms, simply walked past.

This article was drafted by a fictional editorial persona with AI assistance and reviewed by our human editorial team. Sources are cited throughout. How we use AI · Editorial standards