At South Africa’s MeerKAT radio telescope, the search for extraterrestrial technology can run alongside other astronomers’ observations. A review posted on October 5, 2026 reports more than 1.2 million usable pointings processed autonomously since mid-2022. For AI technosignature searches, the scale is changing what researchers can inspect—and what they must filter out.
A separate October 3 review by Breakthrough Listen scientist Steve Croft describes how machine learning is expanding those searches. Both are program reviews posted to arXiv, not announcements of an alien discovery. Croft reports that Listen has found no confirmed technosignatures: evidence of technology beyond Earth.
How AI technosignature searches widen the net
Some models look for familiar signal shapes; others flag observations that differ from their surroundings. That distinction matters when researchers cannot specify every form a distant civilization’s technology might take.
There is already a precedent for revisiting old observations. In January 2023, the SETI Institute reported that a deep-learning search of data from 820 stars uncovered eight previously unidentified signals of interest. Follow-up observations had not recovered them. They were candidates for investigation, not confirmed extraterrestrial transmissions.
The hardest signal to eliminate comes from home
The October 5 history, by David H. E. MacMahon and Daniel J. Czech, recalls BLC1, a signal detected in observations toward Proxima Centauri. Investigators ultimately traced it to terrestrial interference after finding similar signals elsewhere in the data. An apparently promising pattern can survive initial checks and still have an ordinary explanation.
That history also describes systems that search telescope data while other science proceeds, plus an archive with roughly 31 petabytes in use as of March 2026. Preserving data lets researchers return with improved algorithms; deciding what to retain therefore affects what future searches can discover.
A faster search needs its own checks
Croft distinguishes deployed specialist networks from preliminary experiments using multimodal language models to help review candidates. Those experiments remain under evaluation. Tests must measure missed signals as well as rejected interference, especially where live filtering discards raw data.
The practical consequence is a wider search with a continuing burden of proof. In our assessment, AI’s value here depends on making unusual observations testable—not merely making them look extraordinary.
