AI Tool Flags Two Possible Planets in Telescope Data
The preprint has not been peer reviewed, so the two candidate worlds remain unconfirmed.
Updated on Nov. 20, 2026 in Space

Live Poll
Do you trust scientific findings produced by AI tools used by non-experts?
An AI coding tool, Claude Code, was used by Pavel Rabtsevich, an artificial-intelligence specialist in Spain with no formal astronomy education, to search NASA TESS data and identify two possible planets orbiting star TIC 4206066, 116 light years from Earth. The findings are research-stage: Rabtsevich posted them as a Zenodo preprint on 25 September, and they have not been peer reviewed.
Why it matters
The work shows AI tools being pointed at large public telescope archives to surface candidate planets that conventional searches had not flagged. It accelerates a trend in which automated pipelines sift survey data faster than human teams can, though the candidates still require independent confirmation.
The search was run on TESS photometry, which measures the brightness of stars over time, and the AI looked for repeating dips that could indicate orbiting bodies. Two separate periodic dimming signals were found, and the same pattern appears in TESS data from 2018 and 2020, though the preprint has not been peer reviewed and independent confirmation is pending.
The players
Claude Code
An AI coding tool made by Anthropic, used here to run the planet search across the TESS data.
NASA Transiting Exoplanet Survey Satellite (TESS)
NASA's space telescope mission that monitors the brightness of more than 120,000 stars to detect transiting planets.
University of British Columbia
The Vancouver, Canada university where exoplanet researcher Michelle Kunimoto works and whose published planet-detection algorithm was used to validate the signals.
The details
Claude Code used a widely used planet-detection algorithm to scan light curves, which are records of how a star's brightness changes over time, for small, regular dips that could be caused by a planet crossing in front of its star. The AI then used a planet-detection algorithm that exoplanet researcher Michelle Kunimoto of the University of British Columbia had written and made publicly available to validate the signals. Two cycles of dimming were flagged at the star, one roughly every 3 days and another every 11 days, each consistent with a separate orbiting body.
Timeline
• 2018 and 2020: Archival TESS data recorded the same dimming pattern from star TIC 4206066.
• Late 2025 and early 2026: TESS observed the more than 120,000 stars that were later analyzed.
• September 2026: Rabtsevich asked the AI model to search for an undetected planet.
• 25 September 2026: Rabtsevich published his findings as a Zenodo preprint.
The Tech Race
The result places a general-purpose AI coding tool alongside the dedicated planet-detection algorithms developed by astronomers such as Kunimoto at the University of British Columbia. It suggests the competition to find exoplanets in existing survey data is widening to include researchers and tools outside traditional astronomy.
The work is a preprint, so the candidate planets are not yet an established finding for the public or for other researchers to rely on. Its practical effect for readers is that similar AI-assisted searches of public telescope archives are now demonstrated, though no timeline for confirmation has been announced.
The takeaway
The result shows that public telescope archives can be searched by general-purpose AI tools, not only by specialist algorithms. Watch for peer review of the Zenodo preprint and for independent observations that either confirm or reject the two candidate planets.
Further reading
For more on how telescope surveys are being mined for new worlds, explore our Space coverage.
Source note: This article includes information reported by Nature.
Live Poll
Do you trust scientific findings produced by AI tools used by non-experts?

