Amateur Uses AI Agents to Flag Potential Exoplanet in NASA TESS Data

Key points

• Rabtsevich used Anthropic’s Claude Code and OpenAI’s Codex to process over 126,000 stars in TESS data.
• The candidate planet is estimated to be 1.4 times Earth’s size and located 116 light-years away.
• The star’s brightness dips by roughly 0.05% for about two hours every 3.18 days.
• NASA’s TESS team approved a high-cadensity observation window from October 31 to November 26, 2026.
• Rabtsevich ran 74 tests to rule out false positives, including instrumental glitches and binary star systems.

Background

This follows a trend of amateur astronomers using public data for discoveries, such as the recent identification of a 25km impact crater in Quebec. It also aligns with broader efforts to democratize astronomical research through accessible AI tools.

How outlets are covering it

Gizmodo emphasizes the role of AI in democratizing discovery, noting Rabtsevich’s lack of formal training. Scientific American highlights the signal’s presence in older data (2018-2020) and notes that professional teams had previously flagged but rejected the candidate due to insufficient evidence. Cybernews focuses on the technical aspect, detailing how AI agents wrote code and ran tests to verify the signal. LavX News reports on a separate, unrelated discovery involving the Kepler telescope, which should not be confused with the TESS findings.

Why it matters

This case demonstrates how AI tools can enable non-experts to contribute to scientific discovery by processing large datasets. It also highlights the importance of follow-up observations to confirm candidate signals, as false positives remain a significant challenge in exoplanet detection.

What to watch

TESS will observe TIC 4206066 from October 31 to November 26, 2026, measuring brightness every two minutes. Rabtsevich has publicly predicted the timing of dimming events, and the results will determine if the candidate is confirmed as an exoplanet.