The Hook

The Milky Way is 13.6 billion years old. It contains somewhere between 100 billion and 400 billion stars. Around most of those stars, physics insists, orbit planets — a number so large it dissolves into abstraction, like trying to count every grain of sand on every beach on Earth and then multiplying by a thousand. We have confirmed just over 6,000 of them. Six thousand, out of what may be trillions.

Now zoom in. Not to a telescope. Not to a control room. To a single server rack humming in a university building at the University of Warwick, England, running an AI pipeline called RAVEN. In May 2026, that machine did something no human team had managed in years of staring at the same data: it found 118 planets we had missed. Hidden in plain sight. Already recorded. Already measured. Just never seen.

One of them completes an entire year — a full orbit around its sun — in less time than it takes you to sleep, wake up, and eat breakfast. Its year is shorter than 24 hours.

The Deep Dive

To understand what RAVEN actually did, you need to understand the transit method — the primary technique NASA's TESS space telescope uses to hunt planets. TESS stares at stars. Millions of them, simultaneously, measuring their brightness thousands of times over months and years. When a planet passes between its star and our telescope — a transit — it blocks a tiny fraction of that starlight. We're talking about a dimming so subtle it's like watching a streetlight from across a city and trying to detect a single moth fluttering in front of it. That flicker, that whisper of shadow, is how we know a world exists.

TESS has been collecting this data since 2018. It has already confirmed hundreds of exoplanets on its own — a tally that continues to climb as new validations come in. But here's the thing: the data TESS generates is oceanic. Millions of stellar light curves, each one a long, jagged line representing brightness over time. Human astronomers cannot read all of it with the attention it deserves. Patterns get missed. Signals drown in noise. Planets hide.

RAVEN was built to solve exactly that problem. Dr. Marina Lafarga Magro and her team at Warwick trained the AI on known planetary signals, teaching it to recognize the characteristic dip-and-recovery pattern of a transiting world buried inside messy, real-world data. Then they unleashed it on TESS's archive. The results were staggering: 118 confirmed new planets, over 2,000 high-quality planet candidates, and nearly 1,000 of those candidates entirely new — never flagged by any previous survey.

Among the confirmed worlds are planets occupying what astronomers call the "Neptunian desert" — a zone so close to a star that the radiation should theoretically strip away a planet's atmosphere entirely, leaving nothing but a scorched, naked core. Planets aren't supposed to survive there. And yet, there they are. RAVEN found them anyway, orbiting in defiance of our models, demanding that we revise our understanding of how planetary systems form and endure.

Then there are the ultra-short-period planets — the ones that complete a full year in under 24 hours. Picture Wolf 327b, a scorching rocky super-Earth discovered 93 light-years from here, orbiting its host star every 13.7 hours. While you sleep eight hours, work eight hours, and live the remaining eight, that planet has already lapped its sun. Its surface temperature would liquefy rock. Its sky — if it has one — is dominated by a star so close it fills the horizon like a wall of nuclear fire. It is a planet in the loosest sense: a world, yes, but one so alien it shares almost nothing with the ground beneath your feet right now.

What makes RAVEN's achievement genuinely historic isn't just the number of planets found. It's the method. These worlds were always there, encoded in data already beamed down from space, already stored on hard drives, already paid for by taxpayers and processed by engineers. We had the evidence. We lacked the eyes. RAVEN gave us new eyes — and the universe immediately revealed more of itself.

The team validated one of their new candidates by pointing one of the 6.5-meter Magellan telescopes in Chile's Atacama Desert directly at it. The planet — TIC 183374187 b, a hot Jupiter orbiting a star 3,950 light-years away — was exactly where RAVEN predicted. The algorithm was right. Which means the other 2,000 candidates deserve serious attention. Months or years of follow-up work stretch ahead, each confirmation a new world added to humanity's map.

Why It Matters

Here is the vertigo: TESS has surveyed a fraction of the sky. RAVEN combed through that fraction and found over 100 planets humans missed. Scale that across the full celestial sphere — across every star TESS hasn't reached yet, across the targets of the James Webb Space Telescope, across future missions like Canada's proposed POET telescope designed specifically to hunt Earth-sized worlds — and the implication becomes almost too large to hold. We are not running out of planets to find. We are only just learning how to look.

Every confirmed exoplanet sharpens our understanding of Earth's own biography. When we find a planet baking at 2,000 degrees Celsius in a 13-hour orbit, we understand more precisely why Earth's distance from our sun is not an accident to be taken lightly. When we find worlds surviving in the Neptunian desert against all expectation, we learn that planetary resilience is stranger and more tenacious than our models predicted. The universe, it turns out, is more inventive than we are.

RAVEN is not replacing astronomers. It is amplifying them — turning years of potential discovery into months, freeing human scientists to ask deeper questions about the worlds the AI surfaces. It is a collaboration between biological curiosity and machine pattern-recognition, and together they are dismantling the idea that we have already found the interesting stuff.

We have not. We have barely begun. The data is full of shadows. The shadows are full of worlds.

And somewhere in that archive, right now, a planet is transiting its star — dimming a light-curve by 0.1% — waiting for an algorithm to notice what we could not.

Coordinates: The Las Campanas Observatory, Atacama Desert, Chile (29.0146° S, 70.6926° W) — home to the twin Magellan telescopes that confirmed RAVEN's first independently validated planet. On a moonless night, the Atacama sky carries more stars than darkness. Stand there and look up: you are staring at a galaxy that holds trillions of planets, and a machine is learning, right now, to find them all.

An AI called RAVEN just confirmed over 100 planets hiding in NASA data we already had — including worlds where a full year lasts just 13.7 hours — meaning the universe had been quietly leaving us a treasure map and we simply hadn't built the right eyes to read it.

References

  1. Lafarga Magro, M. et al. (2026). AI approach uncovers dozens of hidden planets. University of Warwick Press Release. https://warwick.ac.uk/news/pressreleases/ai-approach-uncovers-dozens-of-hidden-planets/
  2. Astronomy Now (2026). Artificial intelligence uncovers more than 100 new worlds in NASA data. https://astronomynow.com/2026/03/25/artificial-intelligence-uncovers-more-than-100-new-worlds-in-nasa-data/
  3. Phys.org (2024). Wolf 327b: astronomers discover ultra-short-period super-Earth. https://phys.org/news/2024-01-wolf-astronomers