NASA has discovered dozens of fresh craters on Mars by using an AI to analyze orbital image data, a promising new method to study planets in our solar system
Daniel Oberhaus / Wired : Tweets: @daamitt , @brianroemmele , and @wired Tweets: Amit Bhor / @daamitt : I'm hungry and read this as “fresh caterers”. https://twitter.com/... Brian Roemmele / @brianroemmele : Using new AI technologies we will make massive new discoveries in space sciences. Discoveries of fresh craters on Mars will offer new insights on how the planet has a different impact potential then Earth. https://twitter.com/... @wired : Combing through a planet's worth of images looking for the telltale signs of fresh impact is tedious work, but it's exactly the sort of problem that an AI was made to solve. https://www.wired.com/...
Context & Ripple Effects
This is not NASA's first AI-assisted find: back in 2017, a Google neural network surfaced a new planet in the Kepler-90 system from telescope data, establishing the template of machine learning as a discovery instrument rather than just an analysis tool. The crater hunt extends that pattern to planetary surfaces, where the bottleneck is combing through a planet's worth of orbital images for telltale signs of fresh impacts — tedious work that an AI can do at scale.
The method is also proving portable across domains and agencies: researchers recently paired AI with drones to find 303 previously uncharted Nazca Lines in Peru, nearly doubling the mapped total, and ESA astronomers used the AnomalyMatch model to scan 100M Hubble archive cutouts in days. NASA's own AI footprint has been expanding rapidly alongside these efforts, including the open-source Surya solar-weather model built with IBM.
First-order effects
- NASA gains a running census of fresh Martian craters without manual image review, giving planetary scientists a steadier stream of new impact sites to study how Mars' impact rate differs from Earth's.
Second-order effects
- Other space agencies face pressure to apply the same pipeline to their own archives — ESA's AnomalyMatch scan of Hubble data shows the approach already spreading beyond NASA's orbiters.
- Open-source releases like NASA and IBM's Surya lower the barrier for outside researchers to build similar models on public science data, widening who can run discovery-scale searches.
Third-order effects
- If the pattern holds, scientific archives stop being static repositories and become searchable assets, shifting the discovery bottleneck from collecting data to validating what models flag — a structural change in how astronomy and planetary science allocate human review.
- Government science agencies are institutionalizing AI across missions — NASA reported its AI use cases jumping from 18 in 2024 to 420 in 2025 — pointing toward AI becoming standard infrastructure in publicly funded research rather than one-off experiments.
The trend: AI is becoming the default discovery instrument for large scientific image archives, with space agencies racing to turn petabyte-scale datasets into machine-searchable catalogs.