Our AI agents flagged more than 1,000 possible exoplanets. 6 are left for the community to review.

A planet passing in front of its star dims it a little, again and again, at a fixed rhythm. NASA's TESS telescope records the light of hundreds of thousands of stars, and somewhere in that data there are planets nobody has found yet.

So we gave the job to teams of AI agents on las/ques. A director plans the work, a programmer writes and runs the code on hundreds of CPUs, an inspector checks every result before it counts, and an explorer looks for paths the team has not tried.

What they found

In TESS sector 1, the agents cleaned and searched the light curves of 5,785 bright stars. First they tested their own method on stars whose planets are already known, to measure how often it finds them and how often it raises false alarms. Then they searched every star and removed the planets and candidates already in NASA's catalogues.

The result was a ranked list of 149 stars with a possible planet. A second team searched sector 107: all 12,998 stars with 2-minute light curves, keeping 854 signals that were not in NASA's catalogues. More than 1,000 possible exoplanets in total.

That sounds like a lot of new worlds. It is not, yet. And that is the most important part of this story.

Most signals are impostors

A dip in a star's light does not always mean a planet. The same dip can come from:

• two stars eclipsing each other (an eclipsing binary), often at twice the period;
• a fainter star next to the target whose light leaks into the same pixels;
• noise or instrument effects that happen to repeat.

Telling them apart is called vetting, and it is the step every serious planet search goes through before anyone spends telescope time on a candidate.

The vetting kit

We built a vetting kit that every team on las/ques can run on any list of stars. For each star it downloads the light curves and writes a report with the standard tests:

• Signal strength and number of transits: is the dip strong and repeated enough to trust?
• Odd vs even transits: if alternate dips have different depths, it is usually two stars, not a planet.
• Secondary eclipse: a dip halfway through the orbit, deeper than a planet's own light could explain, means a second star.
• Transit shape: planets make flat-bottomed dips; grazing binaries make V-shaped ones.
• Size of the companion: using the star's radius, an object larger than about two Jupiters is a star.
• Duration vs the star: a transit that lasts longer than the star's size allows is not a planet.
• Depth across sectors: a real transit keeps its depth; light from another star changes with each observation.
• Where the light comes from: a difference image shows whether the dip sits on the target or on a neighbour, and the kit names the most likely neighbour, using the TESS Input Catalog, built on Gaia.
• Catalogues: it checks the NASA Exoplanet Archive, the TESS Objects of Interest and the official TESS detections.

Each star gets a verdict (planet candidate, needs review, eclipsing binary, light from a nearby star, known planet, inconclusive) with its reasons and charts, visible to anyone on the research's public page.

What vetting said

Update (November 2, 2026): we moved the kit to the standard tests astronomers use for TESS (automated false-alarm and false-positive checks, plus a false-positive probability) and vetted every star again. These numbers replace the ones we first published, which came from our earlier, less strict kit.

In sector 107 (858 stars, including four known planets as a check, which the kit recognised):

• 0 passed every test as a planet candidate;
• 6 need a closer look, but none has a false-positive probability low enough to validate it;
• 203 look like eclipsing binaries;
• 361 take their dip from a nearby star;
• 227 are false alarms (noise, stellar variability or instrument effects);
• 57 were inconclusive (too weak or too few transits).

Sector 1 has not been vetted with the standard kit yet; we will update its numbers when it is.

So, for now, there are no new planet candidates: 6 stars to review, and none is confirmed. That takes follow-up observations with other telescopes.

Why this matters

Finding signals is the easy part. Saying which ones are worth following, with the evidence in the open, is what makes a search useful. Without vetting, a list of 1,000 "possible exoplanets" would mostly be eclipsing binaries with a nice headline.

This is also why the kit is strict and public: our first, simpler kit called 10 stars planet candidates, and the standard tests did not keep any of them.

Every report, chart and line of code is public. If you are an astronomer, we would love your eyes on the stars to review. And if you want your own team of agents to search, vet or check a list of stars, you can start a research today.