An amateur astronomer harnessed artificial intelligence agents to find a previously undiscovered exoplanet lurking in publicly available data.
Pavel Rabtsevich, a Spain-based product manager and a space science hobbyist, was poring over data collected by NASA’s Transiting Exoplanet Survey Satellite (TESS), an exoplanet-hunting space telescope, when he noticed something odd.
TESS looks for the subtle dimming caused by planets transiting, or passing in front of, their stars. Of the more than 126,000 target stars that Rabtsevich was examining with a custom-built AI algorithm, his machine-learning assistant directed its focus toward one in particular: TIC 4206066. The star was located around 116 light-years away from Earth in the Orion constellation, and the data indicated it dimmed for about two hours every three days.
With his curiosity piqued, Rabtsevich dug through the TESS archives for more data on the same star and found an additional 22 instances of dimming. If the signal were to be confirmed to represent an exoplanet, that world would a super-Earth with a radius around 1.4 times as large as that of our planet. Although such an exoplanet would likely be a rocky world, its three-day orbit of TIC 4206066 would mean it had an estimated surface temperature of 1,000 degrees Fahrenheit (540 degrees Celsius).
On supporting science journalism
If you’re enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
But stars can dim for many other reasons. In this case, for instance, Rabtsevich noticed the dips in TIC 4206066’s light in 2018 tended to align with moments where TESS fired its thrusters, which may have skewed the results. For now, while Rabtsevich hopes this may be a new exoplanet, he acknowledges that all that can be confirmed is a “repeating transitlike signal in open data, with a plausible planetary explanation and alternatives that aren’t yet ruled out.”
“Models agreeing with each other isn’t evidence,” he says. “What I rely on is a signal anyone can measure, calculations anyone can rerun, control samples and rules published before the next [datasets] arrive.”
The search couldn’t have been done without AI, Rabtsevich adds, noting that he used multiple agents and that the models “wrote the code, ran it and checked each other’s work.” Even so, he maintains, “the research itself was mine: the question, the plan, the rules, the computing, the decisions on every candidate and what to publish.”
The potential for at least one exoplanet orbiting TIC 4206066 seems “solid,” says astronomer Greg Laughlin of Yale University, who now works in the AI industry. Rabtsevich’s analysis is “a great example of human-AI teamwork,” he adds: large language models such as Claude and Codex have proven effective at combing astronomical literature for overlooked discoveries.
If there is indeed an exoplanet circling TIC 4206066, Rabtsevich may not actually be the first to have seen it. Pablo Mercader Perez, a Ph.D. student at the Massachusetts Institute of Technology and a member of the TESS team, says the signal was picked up by a detection filter in 2020 but was ultimately rejected because of a lack of conclusive evidence. Since then TESS has conducted many more observations of that part of the sky.
“Thanks to this, we got much more data than we had before for this particular star system, and the signal became a lot clearer,” Mercader Perez says. In terms of statistical significance, “this planet candidate became effectively detectable only in January of this year.”
A deep-learning model developed by Mercader Perez’s team has also flagged the potential exoplanet, he says.
“Independently but also not public yet, a search performed by [astronomer] Michelle Kunimoto using her software called LEO-Vetter had also selected this planet candidate already back in early 2026, which is another way in which it would almost certainly have been alerted in the official list in the next few months,” Mercader Perez says.
Other amateurs, either acting entirely independent of or taking cues from Rabtsevich’s exciting candidate, are now pursuing additional AI-assisted dives into public datasets from TESS and other NASA missions. And some have reported promising early results.
In the meantime, Rabtsevich’s calculations have prompted NASA’s TESS team to reexamine the star once the spacecraft’s orbit around the sun brings the target into position in November. He’s also recruited several other astronomers to look at the star and critique his findings, which include a speculative additional exoplanet in a more distant orbit around the star.
Regardless of who might claim the discovery, “it’s exciting that the parent star has been put on TESS’s high-cadence list, which should elevate the confidence in the detection,” Laughlin says. “There’s every reason to expect that this candidate super-Earth will pass further tests and make it into the official census.”
