Most searches for alien radio signals end the same way. A large pile of candidates gets filtered down to nothing, because the transmitters turn out to be on Earth. A new survey of 33 exoplanet systems using data from China's FAST radio telescope still has one signal left after its filters. That leftover is NBS 260108, a narrowband signal at 1148.4167 MHz that appeared in the telescope beam pointed at a red dwarf called K2-155.

Before the speculation starts: the team does not claim to have found anything artificial from beyond Earth. Their own analysis points to a source closer to home. The way the signal got this far, and the specific reasons it is probably not aliens, explain more about modern technosignature searches than a headline would.

From 139,127 to Two

The work is described in a paper by Zi-Qi Li and 15 co-authors, published in The Astronomical Journal (volume 172, article 180) on August 27, 2026. The paper's main subject is the search method more than any single signal. The authors built a pipeline around a neural network they call the Multi-Scale Wavelet Net, or MSWNet. Around it they added a lightweight parameter estimator, morphology-aware filtering, a signal-to-noise check against the raw data, and a multi-beam anticoincidence veto.

Each stage is aimed at weeding out false positives. Rather than relying only on hard-threshold drift searches, the pipeline treats narrowband detection as wavelet-guided feature extraction followed by endpoint regression. The anticoincidence veto applies a basic SETI test: FAST's L-band receiver observes through 19 beams at once, so the central beam can be trained on the target while the other 18 monitor nearby patches of sky. A signal that appears in several beams is probably terrestrial, while one that appears only in the beam pointed at the target survives another round of scrutiny.

The authors report that on real FAST data, the pipeline recovers representative events from prior analyses. That check matters, because a new detection method should at least find what older methods already found. They say the output is a compact set of "veto-ready" candidates for downstream inspection.

The numbers show how much filtering this takes. According to EarthSky, the survey started with 139,127 initial detections and ended with two. One, toward Kepler-438, had already been identified in an earlier analysis and was ruled out as interference. The other is NBS 260108. The observations were made in 2021. The signal turned up only when the new machine-learning search was run on the archived data.

What Makes NBS 260108 Interesting

K2-155 is a red dwarf about 238 light-years away. ScienceAlert reports that it has three known super-Earths. The outermost, K2-155 d, is about 1.6 times the radius of Earth and sits close to the star's habitable zone. A world like that makes the system an appealing target for a SETI survey.

The signal passed several of the main tests. It is narrow, which natural astrophysical emission rarely is. It drifts in frequency, which is what the Doppler effect would produce for a transmitter on a rotating, orbiting world, although terrestrial interference can sometimes imitate that drift. And it appeared only in the beam pointed at K2-155, not in the other 18 beams. Together, those three properties are what got it through a pipeline that whittled 139,127 detections down to two.

Why It Is Probably Not Aliens

Two details count against it, and both carry weight.

The first is polarization. Both outlets report that NBS 260108 appeared strongly in only one polarization channel of the telescope. ScienceAlert counts this among the clues pointing back toward terrestrial or instrumental interference.

The second detail is more damaging. Similar signals at nearby frequencies appeared in observations of three other stars. A signal from K2-155 d would not be expected to also appear while the telescope is pointed at unrelated systems. When nearly the same frequency keeps showing up across different targets, a source local to Earth or the instrument becomes the more likely explanation.

That is why NBS 260108 remains unclassified, not dismissed. ScienceAlert reports that an Earthly or instrumental origin is considered more likely than an extraterrestrial one, and that the researchers could not identify exactly what produced the signal. EarthSky likewise concludes that terrestrial interference is the more likely source, while noting the source remains unknown. The signal passed the single-beam test but is undercut by the recurrence at other stars.

Why It Matters

It would be easy to read this as one more SETI candidate that didn't pan out. The more useful result is the pipeline. Radio telescopes like FAST produce far more data than people can inspect, and Earth is awash with radio signals of our own making, from satellites to aircraft to navigation systems. A method that can cut 139,127 detections down to two, recover known events from earlier work, and hand people a short list ready for veto tests could make large technosignature surveys more practical. The paper itself calls the workflow readily transferable to other radio surveys and large-scale technosignature searches.

NBS 260108 also shows the method working as intended. It flagged a candidate, and the remaining checks (polarization behaviour and recurrence across targets) pointed toward a likely terrestrial or instrumental origin. A pipeline that only produced clean negatives would be harder to trust than one that brings up close calls and lets people examine them. ScienceAlert makes a similar point: even if it is just interference again, the refinement of SETI processes should make future searches more robust.

The candidate also suggests a next step. Follow-up observations of K2-155 at 1148.4167 MHz could show whether the signal persists. The sources don't say whether any such follow-up is planned.

Timing: the paper was published in the journal in August, and its latest arXiv revision (v4) is dated September 18. Press coverage arrived between September 26 (ScienceAlert) and October 2 (EarthSky). That puts this story outside our usual three-day news window. We're covering it anyway because the method is likely to outlast this particular candidate.

Sources